billion-context 0.1.116 → 0.1.117

package/dist/index.js

@@ -1141,14 +1141,14 @@ var require_util = __commonJS({
         }
         const port = url.port != null ? url.port : url.protocol === "https:" ? 443 : 80;
         let origin = url.origin != null ? url.origin : `${url.protocol || ""}//${url.hostname || ""}:${port}`;
-        let path20 = url.path != null ? url.path : `${url.pathname || ""}${url.search || ""}`;
+        let path22 = url.path != null ? url.path : `${url.pathname || ""}${url.search || ""}`;
         if (origin[origin.length - 1] === "/") {
           origin = origin.slice(0, origin.length - 1);
         }
-        if (path20 && path20[0] !== "/") {
-          path20 = `/${path20}`;
+        if (path22 && path22[0] !== "/") {
+          path22 = `/${path22}`;
         }
-        return new URL(`${origin}${path20}`);
+        return new URL(`${origin}${path22}`);
       }
       if (!isHttpOrHttpsPrefixed(url.origin || url.protocol)) {
         throw new InvalidArgumentError("Invalid URL protocol: the URL must start with `http:` or `https:`.");
@@ -1969,9 +1969,9 @@ var require_diagnostics = __commonJS({
         "undici:client:sendHeaders",
         (evt) => {
           const {
-            request: { method, path: path20, origin }
+            request: { method, path: path22, origin }
           } = evt;
-          debugLog("sending request to %s %s%s", method, origin, path20);
+          debugLog("sending request to %s %s%s", method, origin, path22);
         }
       );
     }
@@ -1989,14 +1989,14 @@ var require_diagnostics = __commonJS({
         "undici:request:headers",
         (evt) => {
           const {
-            request: { method, path: path20, origin },
+            request: { method, path: path22, origin },
             response: { statusCode }
           } = evt;
           debugLog(
             "received response to %s %s%s - HTTP %d",
             method,
             origin,
-            path20,
+            path22,
             statusCode
           );
         }
@@ -2005,23 +2005,23 @@ var require_diagnostics = __commonJS({
         "undici:request:trailers",
         (evt) => {
           const {
-            request: { method, path: path20, origin }
+            request: { method, path: path22, origin }
           } = evt;
-          debugLog("trailers received from %s %s%s", method, origin, path20);
+          debugLog("trailers received from %s %s%s", method, origin, path22);
         }
       );
       diagnosticsChannel.subscribe(
         "undici:request:error",
         (evt) => {
           const {
-            request: { method, path: path20, origin },
+            request: { method, path: path22, origin },
             error
           } = evt;
           debugLog(
             "request to %s %s%s errored - %s",
             method,
             origin,
-            path20,
+            path22,
             error.message
           );
         }
@@ -2136,7 +2136,7 @@ var require_request = __commonJS({
     var kHandler = /* @__PURE__ */ Symbol("handler");
     var Request = class {
       constructor(origin, {
-        path: path20,
+        path: path22,
         method,
         body,
         headers,
@@ -2153,11 +2153,11 @@ var require_request = __commonJS({
         maxRedirections,
         typeOfService
       }, handler) {
-        if (typeof path20 !== "string") {
+        if (typeof path22 !== "string") {
           throw new InvalidArgumentError("path must be a string");
-        } else if (path20[0] !== "/" && !(path20.startsWith("http://") || path20.startsWith("https://")) && method !== "CONNECT") {
+        } else if (path22[0] !== "/" && !(path22.startsWith("http://") || path22.startsWith("https://")) && method !== "CONNECT") {
           throw new InvalidArgumentError("path must be an absolute URL or start with a slash");
-        } else if (invalidPathRegex.test(path20)) {
+        } else if (invalidPathRegex.test(path22)) {
           throw new InvalidArgumentError("invalid request path");
         }
         if (typeof method !== "string") {
@@ -2232,7 +2232,7 @@ var require_request = __commonJS({
         this.completed = false;
         this.aborted = false;
         this.upgrade = upgrade || null;
-        this.path = query ? serializePathWithQuery(path20, query) : path20;
+        this.path = query ? serializePathWithQuery(path22, query) : path22;
         this.origin = origin;
         this.protocol = getProtocolFromUrlString(origin);
         this.idempotent = idempotent == null ? method === "HEAD" || method === "GET" : idempotent;
@@ -7415,7 +7415,7 @@ var require_client_h1 = __commonJS({
       return method !== "GET" && method !== "HEAD" && method !== "OPTIONS" && method !== "TRACE" && method !== "CONNECT";
     }
     function writeH1(client, request) {
-      const { method, path: path20, host, upgrade, blocking, reset } = request;
+      const { method, path: path22, host, upgrade, blocking, reset } = request;
       let { body, headers, contentLength } = request;
       const expectsPayload = method === "PUT" || method === "POST" || method === "PATCH" || method === "QUERY" || method === "PROPFIND" || method === "PROPPATCH";
       if (util.isFormDataLike(body)) {
@@ -7493,7 +7493,7 @@ var require_client_h1 = __commonJS({
       if (socket.setTypeOfService) {
         socket.setTypeOfService(request.typeOfService);
       }
-      let header = `${method} ${path20} HTTP/1.1\r
+      let header = `${method} ${path22} HTTP/1.1\r
 `;
       if (typeof host === "string") {
         header += `host: ${host}\r
@@ -8146,7 +8146,7 @@ var require_client_h2 = __commonJS({
     function writeH2(client, request) {
       const requestTimeout = request.bodyTimeout ?? client[kBodyTimeout];
       const session = client[kHTTP2Session];
-      const { method, path: path20, host, upgrade, expectContinue, signal, protocol, headers: reqHeaders } = request;
+      const { method, path: path22, host, upgrade, expectContinue, signal, protocol, headers: reqHeaders } = request;
       let { body } = request;
       if (upgrade != null && upgrade !== "websocket") {
         util.errorRequest(client, request, new InvalidArgumentError(`Custom upgrade "${upgrade}" not supported over HTTP/2`));
@@ -8214,7 +8214,7 @@ var require_client_h2 = __commonJS({
           }
           headers[HTTP2_HEADER_METHOD] = "CONNECT";
           headers[HTTP2_HEADER_PROTOCOL] = "websocket";
-          headers[HTTP2_HEADER_PATH] = path20;
+          headers[HTTP2_HEADER_PATH] = path22;
           if (protocol === "ws:" || protocol === "wss:") {
             headers[HTTP2_HEADER_SCHEME] = protocol === "ws:" ? "http" : "https";
           } else {
@@ -8255,7 +8255,7 @@ var require_client_h2 = __commonJS({
         stream2.setTimeout(requestTimeout);
         return true;
       }
-      headers[HTTP2_HEADER_PATH] = path20;
+      headers[HTTP2_HEADER_PATH] = path22;
       headers[HTTP2_HEADER_SCHEME] = protocol === "http:" ? "http" : "https";
       const expectsPayload = method === "PUT" || method === "POST" || method === "PATCH";
       if (body && typeof body.read === "function") {
@@ -10598,10 +10598,10 @@ var require_proxy_agent = __commonJS({
         };
         const {
           origin,
-          path: path20 = "/",
+          path: path22 = "/",
           headers = {}
         } = opts;
-        opts.path = origin + path20;
+        opts.path = origin + path22;
         if (!("host" in headers) && !("Host" in headers)) {
           const { host } = new URL(origin);
           headers.host = host;
@@ -12684,20 +12684,20 @@ var require_mock_utils = __commonJS({
       }
       return normalizedQp;
     }
-    function safeUrl(path20) {
-      if (typeof path20 !== "string") {
-        return path20;
+    function safeUrl(path22) {
+      if (typeof path22 !== "string") {
+        return path22;
       }
-      const pathSegments = path20.split("?", 3);
+      const pathSegments = path22.split("?", 3);
       if (pathSegments.length !== 2) {
-        return path20;
+        return path22;
       }
       const qp = new URLSearchParams(pathSegments.pop());
       qp.sort();
       return [...pathSegments, qp.toString()].join("?");
     }
-    function matchKey(mockDispatch2, { path: path20, method, body, headers }) {
-      const pathMatch = matchValue(mockDispatch2.path, path20);
+    function matchKey(mockDispatch2, { path: path22, method, body, headers }) {
+      const pathMatch = matchValue(mockDispatch2.path, path22);
       const methodMatch = matchValue(mockDispatch2.method, method);
       const bodyMatch = typeof mockDispatch2.body !== "undefined" ? matchValue(mockDispatch2.body, body) : true;
       const headersMatch = matchHeaders(mockDispatch2, headers);
@@ -12722,8 +12722,8 @@ var require_mock_utils = __commonJS({
       const basePath = key.query ? serializePathWithQuery(key.path, key.query) : key.path;
       const resolvedPath = typeof basePath === "string" ? safeUrl(basePath) : basePath;
       const resolvedPathWithoutTrailingSlash = removeTrailingSlash(resolvedPath);
-      let matchedMockDispatches = mockDispatches.filter(({ consumed }) => !consumed).filter(({ path: path20, ignoreTrailingSlash }) => {
-        return ignoreTrailingSlash ? matchValue(removeTrailingSlash(safeUrl(path20)), resolvedPathWithoutTrailingSlash) : matchValue(safeUrl(path20), resolvedPath);
+      let matchedMockDispatches = mockDispatches.filter(({ consumed }) => !consumed).filter(({ path: path22, ignoreTrailingSlash }) => {
+        return ignoreTrailingSlash ? matchValue(removeTrailingSlash(safeUrl(path22)), resolvedPathWithoutTrailingSlash) : matchValue(safeUrl(path22), resolvedPath);
       });
       if (matchedMockDispatches.length === 0) {
         throw new MockNotMatchedError(`Mock dispatch not matched for path '${resolvedPath}'`);
@@ -12762,19 +12762,19 @@ var require_mock_utils = __commonJS({
         mockDispatches.splice(index, 1);
       }
     }
-    function removeTrailingSlash(path20) {
-      while (path20.endsWith("/")) {
-        path20 = path20.slice(0, -1);
+    function removeTrailingSlash(path22) {
+      while (path22.endsWith("/")) {
+        path22 = path22.slice(0, -1);
       }
-      if (path20.length === 0) {
-        path20 = "/";
+      if (path22.length === 0) {
+        path22 = "/";
       }
-      return path20;
+      return path22;
     }
     function buildKey(opts) {
-      const { path: path20, method, body, headers, query } = opts;
+      const { path: path22, method, body, headers, query } = opts;
       return {
-        path: path20,
+        path: path22,
         method,
         body,
         headers,
@@ -13464,10 +13464,10 @@ var require_pending_interceptors_formatter = __commonJS({
       }
       format(pendingInterceptors) {
         const withPrettyHeaders = pendingInterceptors.map(
-          ({ method, path: path20, data: { statusCode }, persist, times, timesInvoked, origin }) => ({
+          ({ method, path: path22, data: { statusCode }, persist, times, timesInvoked, origin }) => ({
             Method: method,
             Origin: origin,
-            Path: path20,
+            Path: path22,
             "Status code": statusCode,
             Persistent: persist ? PERSISTENT : NOT_PERSISTENT,
             Invocations: timesInvoked,
@@ -13549,9 +13549,9 @@ var require_mock_agent = __commonJS({
         const acceptNonStandardSearchParameters = this[kMockAgentAcceptsNonStandardSearchParameters];
         const dispatchOpts = { ...opts };
         if (acceptNonStandardSearchParameters && dispatchOpts.path) {
-          const [path20, searchParams] = dispatchOpts.path.split("?");
+          const [path22, searchParams] = dispatchOpts.path.split("?");
           const normalizedSearchParams = normalizeSearchParams(searchParams, acceptNonStandardSearchParameters);
-          dispatchOpts.path = `${path20}?${normalizedSearchParams}`;
+          dispatchOpts.path = `${path22}?${normalizedSearchParams}`;
         }
         return this[kAgent].dispatch(dispatchOpts, handler);
       }
@@ -13952,12 +13952,12 @@ var require_snapshot_recorder = __commonJS({
        * @return {Promise<void>} - Resolves when snapshots are loaded
        */
       async loadSnapshots(filePath) {
-        const path20 = filePath || this.#snapshotPath;
-        if (!path20) {
+        const path22 = filePath || this.#snapshotPath;
+        if (!path22) {
           throw new InvalidArgumentError("Snapshot path is required");
         }
         try {
-          const data = await readFile4(resolve(path20), "utf8");
+          const data = await readFile4(resolve(path22), "utf8");
           const parsed = JSON.parse(data);
           if (Array.isArray(parsed)) {
             this.#snapshots.clear();
@@ -13971,7 +13971,7 @@ var require_snapshot_recorder = __commonJS({
           if (error.code === "ENOENT") {
             this.#snapshots.clear();
           } else {
-            throw new UndiciError(`Failed to load snapshots from ${path20}`, { cause: error });
+            throw new UndiciError(`Failed to load snapshots from ${path22}`, { cause: error });
           }
         }
       }
@@ -13982,11 +13982,11 @@ var require_snapshot_recorder = __commonJS({
        * @returns {Promise<void>} - Resolves when snapshots are saved
        */
       async saveSnapshots(filePath) {
-        const path20 = filePath || this.#snapshotPath;
-        if (!path20) {
+        const path22 = filePath || this.#snapshotPath;
+        if (!path22) {
           throw new InvalidArgumentError("Snapshot path is required");
         }
-        const resolvedPath = resolve(path20);
+        const resolvedPath = resolve(path22);
         await mkdir3(dirname5(resolvedPath), { recursive: true });
         const data = Array.from(this.#snapshots.entries()).map(([hash, snapshot]) => ({
           hash,
@@ -14618,15 +14618,15 @@ var require_redirect_handler = __commonJS({
           return;
         }
         const { origin, pathname, search } = util.parseURL(new URL(this.location, this.opts.origin && new URL(this.opts.path, this.opts.origin)));
-        const path20 = search ? `${pathname}${search}` : pathname;
-        const redirectUrlString = `${origin}${path20}`;
+        const path22 = search ? `${pathname}${search}` : pathname;
+        const redirectUrlString = `${origin}${path22}`;
         for (const historyUrl of this.history) {
           if (historyUrl.toString() === redirectUrlString) {
             throw new InvalidArgumentError(`Redirect loop detected. Cannot redirect to ${origin}. This typically happens when using a Client or Pool with cross-origin redirects. Use an Agent for cross-origin redirects.`);
           }
         }
         this.opts.headers = cleanRequestHeaders(this.opts.headers, statusCode === 303, this.opts.origin !== origin);
-        this.opts.path = path20;
+        this.opts.path = path22;
         this.opts.origin = origin;
         this.opts.query = null;
       }
@@ -16395,10 +16395,10 @@ var require_cache_handler = __commonJS({
       }
       return locationUrl.pathname + locationUrl.search;
     }
-    function deleteCachedUri(store, cacheKey, path20) {
+    function deleteCachedUri(store, cacheKey, path22) {
       deleteCachedValue(store, {
         ...cacheKey,
-        path: path20
+        path: path22
       });
       for (let i = 0; i < util.safeHTTPMethods.length; i++) {
         const method = util.safeHTTPMethods[i];
@@ -16406,7 +16406,7 @@ var require_cache_handler = __commonJS({
           deleteCachedValue(store, {
             ...cacheKey,
             method,
-            path: path20
+            path: path22
           });
         }
       }
@@ -16417,9 +16417,9 @@ var require_cache_handler = __commonJS({
       }
       const values = Array.isArray(headerValue3) ? headerValue3 : [headerValue3];
       for (let i = 0; i < values.length; i++) {
-        const path20 = getSameOriginPath(cacheKey, values[i]);
-        if (path20 !== void 0) {
-          deleteCachedUri(store, cacheKey, path20);
+        const path22 = getSameOriginPath(cacheKey, values[i]);
+        if (path22 !== void 0) {
+          deleteCachedUri(store, cacheKey, path22);
         }
       }
     }
@@ -21297,11 +21297,11 @@ var require_fetch = __commonJS({
       function dispatch({ body }) {
         const url = requestCurrentURL(request);
         const agent = fetchParams.controller.dispatcher;
-        const path20 = url.pathname + url.search;
+        const path22 = url.pathname + url.search;
         const hasTrailingQuestionMark = url.search.length === 0 && url.href[url.href.length - url.hash.length - 1] === "?";
         return new Promise((resolve, reject) => agent.dispatch(
           {
-            path: hasTrailingQuestionMark ? `${path20}?` : path20,
+            path: hasTrailingQuestionMark ? `${path22}?` : path22,
             origin: url.origin,
             method: request.method,
             body: agent.isMockActive ? request.body && (request.body.source || request.body.stream) : body,
@@ -22248,9 +22248,9 @@ var require_util4 = __commonJS({
         }
       }
     }
-    function validateCookiePath(path20) {
-      for (let i = 0; i < path20.length; ++i) {
-        const code = path20.charCodeAt(i);
+    function validateCookiePath(path22) {
+      for (let i = 0; i < path22.length; ++i) {
+        const code = path22.charCodeAt(i);
         if (code < 32 || // exclude CTLs (0-31)
         code > 126 || // exclude DEL and non-ascii
         code === 59) {
@@ -25487,11 +25487,11 @@ var require_undici = __commonJS({
           if (typeof opts.path !== "string") {
             throw new InvalidArgumentError("invalid opts.path");
           }
-          let path20 = opts.path;
+          let path22 = opts.path;
           if (!opts.path.startsWith("/")) {
-            path20 = `/${path20}`;
+            path22 = `/${path22}`;
           }
-          url = new URL(util.parseOrigin(url).origin + path20);
+          url = new URL(util.parseOrigin(url).origin + path22);
         } else {
           if (!opts) {
             opts = typeof url === "object" ? url : {};
@@ -48274,6 +48274,11 @@ function stateDir() {
 function defaultLogFile() {
   return path2.join(stateDir(), "bili.log");
 }
+function dumpsDir() {
+  const env = process.env.ACP_DUMP_DIR;
+  if (env && env.length > 0) return env;
+  return path2.join(stateDir(), "dumps");
+}
 function caDir() {
   return path2.join(dataDir(), "ca");
 }
@@ -49155,13 +49160,13 @@ function applyCompatRoles(body, protocol, roles) {
 }
 
 // src/config.ts
-function safeReadJson(path20) {
+function safeReadJson(path22) {
   try {
-    const raw = readFileSync2(path20, "utf8").replace(/^\uFEFF/, "");
+    const raw = readFileSync2(path22, "utf8").replace(/^\uFEFF/, "");
     return JSON.parse(raw);
   } catch (e) {
     if (e.code !== "ENOENT") {
-      log("error", `[acp-config] failed to parse ${path20}: ${String(e)}`);
+      log("error", `[acp-config] failed to parse ${path22}: ${String(e)}`);
     }
     return void 0;
   }
@@ -49177,12 +49182,14 @@ var CONTEXT_LIMIT_TABLE = [
   { match: /^gemini-1\.5/i, limit: 1e6 },
   { match: /^glm-4\.6/i, limit: 128e3 },
   { match: /^glm-5/i, limit: 1e6 },
-  { match: /^glm-/i, limit: 128e3 },
-  { match: /^deepseek/i, limit: 128e3 },
+  { match: /^glm-/i, limit: 2e5 },
+  // DeepSeek: flagship line (chat/reasoner/v4*/flash) is 1M on models.dev; only legacy r1/v3/ocr stay ~128k (#852).
+  { match: /^deepseek-(r1|v3|ocr)/i, limit: 128e3 },
+  { match: /^deepseek/i, limit: 1e6 },
   { match: /^minimax/i, limit: 204800 },
-  { match: /^qwen/i, limit: 128e3 },
-  { match: /^kimi/i, limit: 128e3 },
-  { match: /^llama-/i, limit: 128e3 }
+  { match: /^qwen/i, limit: 2e5 },
+  { match: /^kimi/i, limit: 2e5 },
+  { match: /^llama-/i, limit: 2e5 }
 ];
 function lookupContextLimit(model) {
   if (!model) return void 0;
@@ -49557,6 +49564,7 @@ function rejectLegacyRoute(key, value) {
 // src/server.ts
 import http from "http";
 import fs9 from "fs";
+import path16 from "path";
 import { createHash as createHash8, randomUUID as randomUUID4 } from "crypto";
 
 // src/compress-settings.ts
@@ -49813,7 +49821,7 @@ import { existsSync as existsSync2, statSync as statSync2 } from "fs";
 import path5 from "path";
 
 // src/registry-snapshot.json
-var registry_snapshot_default = { fetchedAt: "2026-08-24T10:39:50.602Z", count: 355, models: { "tencent/hy3-preview": { id: "tencent/hy3-preview", name: "Hy3 preview", description: "Tencent Hy reasoning model for coding, instruction following, and agent tasks", family: "Hy", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-04-20", last_updated: "2026-04-20", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 64e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/tencent/Hy3-preview" }], benchmarks: [{ name: "SWE-Bench Verified", score: 74.4, metric: "resolved", source: "https://huggingface.co/tencent/Hy3-preview" }] }, "tencent/hy3": { id: "tencent/hy3", name: "Hy3", description: "Tencent Hy reasoning model for coding, instruction following, and agent tasks", family: "Hy", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-07-06", last_updated: "2026-07-06", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 64e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/tencent/Hy3" }], benchmarks: [{ name: "SWE-Bench Verified", score: 78, metric: "resolved", source: "https://huggingface.co/tencent/Hy3" }] }, "nvidia/llama-3.3-nemotron-super-49b-v1": { id: "nvidia/llama-3.3-nemotron-super-49b-v1", name: "Llama 3.3 Nemotron Super 49B v1", description: "Nemotron model for efficient reasoning, coding, and specialized AI agents", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-04-07", last_updated: "2025-04-07", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 } }, "nvidia/nemotron-3-nano-30b-a3b": { id: "nvidia/nemotron-3-nano-30b-a3b", name: "Nemotron 3 Nano 30B A3B", description: "Small Nemotron 3 MoE for efficient coding, math, and long-context agents", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-12-15", last_updated: "2025-12-15", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 } }, "nvidia/nemotron-3.5-content-safety": { id: "nvidia/nemotron-3.5-content-safety", name: "Nemotron 3.5 Content Safety", description: "Safety model for policy screening, moderation, and risk-aware routing workflows", family: "nemotron", attachment: true, reasoning: true, tool_call: false, temperature: true, release_date: "2026-06-04", last_updated: "2026-06-04", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "nvidia/nemotron-cascade-2-30b-a3b": { id: "nvidia/nemotron-cascade-2-30b-a3b", name: "Nemotron Cascade 2 30B A3B", description: "Nemotron model for efficient reasoning, coding, and specialized AI agents", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-03-24", last_updated: "2026-04-09", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 32768 } }, "nvidia/nemotron-3-content-safety": { id: "nvidia/nemotron-3-content-safety", name: "Nemotron 3 Content Safety", description: "Safety model for policy screening, moderation, and risk-aware routing workflows", family: "nemotron", attachment: false, reasoning: false, tool_call: false, temperature: false, release_date: "2026-04-16", last_updated: "2026-04-16", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 } }, "nvidia/llama-3.1-nemotron-70b-instruct": { id: "nvidia/llama-3.1-nemotron-70b-instruct", name: "Llama 3.1 Nemotron 70B Instruct", description: "Nemotron model for efficient reasoning, coding, and specialized AI agents", family: "nemotron", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2025-04-15", last_updated: "2025-04-15", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "nvidia/nemotron-nano-12b-v2-vl": { id: "nvidia/nemotron-nano-12b-v2-vl", name: "Nemotron Nano 12B v2 VL", description: "Nemotron multimodal model for visual reasoning and agentic AI workflows", family: "nemotron", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2025-10-28", last_updated: "2025-10-28", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 } }, "nvidia/nemotron-voicechat": { id: "nvidia/nemotron-voicechat", name: "Nemotron VoiceChat", description: "Nemotron multimodal model for visual reasoning and agentic AI workflows", family: "nemotron", attachment: true, reasoning: false, tool_call: true, temperature: true, release_date: "2026-03-16", last_updated: "2026-03-16", modalities: { input: ["text", "audio"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "nvidia/llama-nemotron-rerank-vl-1b-v2": { id: "nvidia/llama-nemotron-rerank-vl-1b-v2", name: "Llama Nemotron Rerank VL 1B v2", description: "Reranking model for improving retrieval quality in search and recommendation systems", family: "nemotron", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2026-03-31", last_updated: "2026-03-31", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 } }, "nvidia/llama-3.3-nemotron-super-49b-v1.5": { id: "nvidia/llama-3.3-nemotron-super-49b-v1.5", name: "Llama 3.3 Nemotron Super 49B v1.5", description: "Nemotron model for efficient reasoning, coding, and specialized AI agents", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-07-25", last_updated: "2025-07-25", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 } }, "nvidia/nemotron-3-super-120b-a12b": { id: "nvidia/nemotron-3-super-120b-a12b", name: "Nemotron 3 Super 120B A12B", description: "Nemotron middle tier for collaborative agents and high-volume reasoning workloads", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-03-11", last_updated: "2026-03-11", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 } }, "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning": { id: "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning", name: "Nemotron 3 Nano Omni 30B A3B Reasoning", description: "Open Nemotron omni model combining reasoning with text, vision, and audio", family: "nemotron", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-04-28", last_updated: "2026-04-28", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 65536 } }, "nvidia/nemotron-3.5-lightning": { id: "nvidia/nemotron-3.5-lightning", name: "Nemotron 3.5 Lightning 30B A3B", description: "Fast NVIDIA Nemotron MoE for reliable agentic tasks across enterprise workloads", family: "nemotron", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-11", last_updated: "2026-08-11", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 } }, "nvidia/nemotron-content-safety-reasoning-4b": { id: "nvidia/nemotron-content-safety-reasoning-4b", name: "Nemotron Content Safety Reasoning 4B", description: "Safety model for policy screening, moderation, and risk-aware routing workflows", family: "nemotron", attachment: false, reasoning: true, tool_call: false, temperature: false, release_date: "2026-01-22", last_updated: "2026-01-22", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 } }, "nvidia/nemotron-3-ultra-550b-a55b": { id: "nvidia/nemotron-3-ultra-550b-a55b", name: "Nemotron 3 Ultra 550B A55B", description: "Largest Nemotron 3 model for maximum open-weight reasoning and agent accuracy", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-04", last_updated: "2026-06-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Verified", score: 70.7, metric: "resolved", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "SWE-Bench Multilingual", score: 67.7, metric: "resolve rate", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "Terminal-Bench", score: 56.4, metric: "success rate", version: "2.1", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "GPQA", score: 87, metric: "accuracy", variant: "no tools", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "Humanity's Last Exam", score: 26.7, metric: "accuracy", variant: "no tools", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "Humanity's Last Exam", score: 37.4, metric: "accuracy", variant: "with tools", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "LiveCodeBench", score: 89, metric: "pass@1", version: "v6", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "MMLU-Pro", score: 86.8, metric: "accuracy", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "BrowseComp", score: 44.4, metric: "accuracy", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "IFBench", score: 81.7, metric: "accuracy", variant: "prompt loose", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "GDPval", score: 46.7, metric: "wins or ties", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }] }, "nvidia/mistral-nemotron": { id: "nvidia/mistral-nemotron", name: "Mistral Nemotron", description: "Mistral model for multilingual chat, reasoning, and tool-assisted workflows", family: "nemotron", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2025-06-11", last_updated: "2025-06-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "nvidia/llama-3.1-nemotron-safety-guard-8b-v3": { id: "nvidia/llama-3.1-nemotron-safety-guard-8b-v3", name: "Llama 3.1 Nemotron Safety Guard 8B v3", description: "Safety model for policy screening, moderation, and risk-aware routing workflows", family: "nemotron", attachment: false, reasoning: false, tool_call: false, temperature: false, release_date: "2025-10-28", last_updated: "2025-10-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 } }, "nvidia/nemotron-nano-9b-v2": { id: "nvidia/nemotron-nano-9b-v2", name: "Nemotron Nano 9B v2", description: "Compact Nemotron model for efficient reasoning and deployable AI agents", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-08-18", last_updated: "2025-08-18", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 } }, "nvidia/llama-3.1-nemotron-ultra-253b": { id: "nvidia/llama-3.1-nemotron-ultra-253b", name: "Llama 3.1 Nemotron Ultra 253B", description: "Flagship Nemotron model for high-throughput reasoning and complex agents", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-04-07", last_updated: "2025-04-07", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "nvidia/nemotron-mini-4b-instruct": { id: "nvidia/nemotron-mini-4b-instruct", name: "Nemotron Mini 4B Instruct", description: "Compact Nemotron model for efficient reasoning and deployable AI agents", family: "nemotron", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2024-08-21", last_updated: "2024-08-26", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "nvidia/llama-nemotron-embed-vl-1b-v2": { id: "nvidia/llama-nemotron-embed-vl-1b-v2", name: "Llama Nemotron Embed VL 1B v2", description: "Embedding model for semantic search, retrieval, clustering, and ranking pipelines", family: "nemotron", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2026-02-10", last_updated: "2026-02-10", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 32768, output: 2048 } }, "aisingapore/gemma-sea-lion-v4-27b-it": { id: "aisingapore/gemma-sea-lion-v4-27b-it", name: "Gemma-SEA-LION-v4-27B-IT", description: "Gemma 3 27B tuned by AI Singapore for Southeast Asian languages and instruction following", family: "gemma", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2025-09-23", last_updated: "2025-09-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT" }] }, "microsoft/phi-4-mini": { id: "microsoft/phi-4-mini", name: "Phi-4-mini", description: "Compact Microsoft instruction model tuned for efficient coding assistance, reasoning, and low-latency agent tasks", family: "phi", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2023-10", release_date: "2024-12-11", last_updated: "2024-12-11", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 }, links: [{ label: "Weights", url: "https://huggingface.co/microsoft/Phi-4-mini-instruct", type: "weights" }], benchmarks: [{ name: "MMLU", score: 67.3, metric: "accuracy", source: "https://huggingface.co/microsoft/Phi-4-mini-instruct/resolve/main/README.md" }] }, "microsoft/mai-code-1-flash": { id: "microsoft/mai-code-1-flash", name: "MAI-Code-1-Flash", description: "Microsoft coding model built for fast, efficient assistance in everyday developer workflows", family: "mai", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-12", release_date: "2026-06-02", last_updated: "2026-06-08", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 128e3 }, links: [{ label: "Model card", url: "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", type: "model_card" }, { label: "Announcement", url: "https://microsoft.ai/news/introducingmai-code-1-flash/", type: "announcement" }], benchmarks: [{ name: "SWE-Bench Pro", score: 51.2, metric: "resolve rate", harness: "GitHub Copilot", source: "https://microsoft.ai/news/introducingmai-code-1-flash/", date: "2026-06-02" }, { name: "SWE-Bench Verified", score: 71.6, metric: "resolved", source: "https://llm-stats.com/benchmarks/swe-bench-verified" }, { name: "Terminal-Bench", score: 54.8, metric: "success rate", version: "2.0", source: "https://llm-stats.com/benchmarks/terminal-bench-2" }, { name: "GPQA Diamond", score: 84.6, metric: "accuracy", source: "https://llm-stats.com/benchmarks/gpqa" }] }, "microsoft/mai-code-1.1-flash": { id: "microsoft/mai-code-1.1-flash", name: "MAI-Code-1.1-Flash", description: "Microsoft coding model with native vision support, optimized for fast and efficient software development", family: "mai", attachment: true, reasoning: true, tool_call: true, structured_output: true, release_date: "2026-08-11", last_updated: "2026-08-11", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 128e3 }, links: [{ label: "Announcement", url: "https://microsoft.ai/news/mai-code-1-1-flash-br-better-faster-at-a-quarter-of-the-cost/", type: "announcement" }] }, "deepseek/deepseek-r1-distill-qwen-32b": { id: "deepseek/deepseek-r1-distill-qwen-32b", name: "DeepSeek-R1-Distill-Qwen-32B", description: "R1 reasoning distilled into Qwen 2.5 32B for efficient open-weight step-by-step problem solving", family: "deepseek-thinking", attachment: false, reasoning: true, tool_call: false, temperature: true, release_date: "2025-01-20", last_updated: "2025-01-20", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B" }] }, "deepseek/deepseek-r1": { id: "deepseek/deepseek-r1", name: "DeepSeek-R1", description: "Classic open reasoning model for transparent math, coding, and deliberate problem solving", family: "deepseek-thinking", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-07", release_date: "2025-01-20", last_updated: "2025-05-29", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-R1" }], benchmarks: [{ name: "Aider Polyglot", score: 56.9, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-01-20" }, { name: "Artificial Analysis Coding Index", score: 15.9, metric: "index", source: "https://openrouter.ai/deepseek/deepseek-r1/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 35.7, metric: "percent correct", source: "https://openrouter.ai/deepseek/deepseek-r1/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 6.1, metric: "success rate", source: "https://openrouter.ai/deepseek/deepseek-r1/benchmarks", date: "2026-03-11" }] }, "deepseek/deepseek-v3-0324": { id: "deepseek/deepseek-v3-0324", name: "DeepSeek V3 0324", description: "March 2025 checkpoint of DeepSeek-V3 with improved reasoning and coding", family: "deepseek", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2025-03-24", last_updated: "2025-03-24", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 163840, output: 163840 }, weights: [{ label: "Model weights", url: "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324", format: "safetensors" }] }, "deepseek/deepseek-v4-flash": { id: "deepseek/deepseek-v4-flash", name: "DeepSeek V4 Flash", description: "Fast DeepSeek V4 lane for economical reasoning, coding, and long-context work", family: "deepseek-flash", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-05", release_date: "2026-04-24", last_updated: "2026-04-24", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" }], benchmarks: [{ name: "SWE-Bench Verified", score: 79, metric: "resolved", source: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" }] }, "deepseek/deepseek-v4-pro-0813": { id: "deepseek/deepseek-v4-pro-0813", name: "DeepSeek V4 Pro 0813", description: "DeepSeek V4 Pro snapshot with million-token context and support for thinking and non-thinking modes", family: "deepseek-thinking", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-12", last_updated: "2026-08-22", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, license: "MIT", weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" }] }, "deepseek/deepseek-v3.1": { id: "deepseek/deepseek-v3.1", name: "DeepSeek-V3.1", description: "Hybrid-reasoning DeepSeek model with thinking and non-thinking modes", family: "deepseek", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-08-21", last_updated: "2025-08-21", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, license: "MIT License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V3.1" }] }, "deepseek/deepseek-v4-pro": { id: "deepseek/deepseek-v4-pro", name: "DeepSeek V4 Pro", description: "Open MoE flagship with million-token context for coding and long agent runs", family: "deepseek-thinking", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-05", release_date: "2026-04-24", last_updated: "2026-04-24", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" }], benchmarks: [{ name: "SWE-Bench Verified", score: 80.6, metric: "resolved", source: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" }, { name: "Artificial Analysis Coding Agent Index", score: 50.1, metric: "average pass@1", harness: "Claude Code", variant: "high", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 67.8, metric: "pass@1", harness: "Claude Code", variant: "high", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 18, metric: "pass@1", harness: "Claude Code", variant: "high", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 64.7, metric: "pass@1", harness: "Claude Code", variant: "high", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }] }, "deepseek/deepseek-v3": { id: "deepseek/deepseek-v3", name: "DeepSeek-V3", description: "Open DeepSeek MoE chat model for coding, math, and general reasoning", family: "deepseek", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2024-12-26", last_updated: "2024-12-26", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, license: "DeepSeek Model License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V3" }] }, "deepseek/deepseek-v3.2": { id: "deepseek/deepseek-v3.2", name: "DeepSeek V3.2", description: "Hybrid-reasoning DeepSeek model with thinking and non-thinking modes, sparse attention, and tool-use", family: "deepseek", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-07", release_date: "2025-12-01", last_updated: "2025-12-01", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 64e3 }, license: "MIT License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V3.2" }] }, "deepseek/deepseek-chat": { id: "deepseek/deepseek-chat", name: "DeepSeek Chat", description: "DeepSeek chat model for instruction following, coding, and analysis", family: "deepseek", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-09", release_date: "2025-12-01", last_updated: "2026-02-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V3.2" }], benchmarks: [{ name: "Aider Polyglot", score: 70.2, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-10-03" }] }, "deepseek/deepseek-reasoner": { id: "deepseek/deepseek-reasoner", name: "DeepSeek Reasoner", description: "DeepSeek reasoning model for multi-step analysis, math, coding, and tools", family: "deepseek-thinking", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-09", release_date: "2025-12-01", last_updated: "2026-02-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V3.2" }], benchmarks: [{ name: "Aider Polyglot", score: 74.2, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-10-03" }] }, "deepseek/deepseek-ocr-2": { id: "deepseek/deepseek-ocr-2", name: "DeepSeek OCR 2", description: "High-accuracy OCR model for extracting text from documents, screenshots, receipts, and natural scenes", attachment: true, reasoning: false, tool_call: false, release_date: "2026-01-27", last_updated: "2026-01-27", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 8192, output: 8192 } }, "deepseek/deepseek-v4-flash-0731": { id: "deepseek/deepseek-v4-flash-0731", name: "DeepSeek V4 Flash 0731", description: "Official DeepSeek V4 Flash release with enhanced agentic capabilities and integrated DSpark speculative decoding", family: "deepseek-flash", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-05", release_date: "2026-07-31", last_updated: "2026-07-31", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, license: "MIT", weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731" }], benchmarks: [{ name: "Terminal-Bench", score: 82.7, metric: "pass@1", variant: "max", version: "2.1", source: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731" }, { name: "NL2Repo", score: 54.2, metric: "resolve rate", harness: "DeepSeek Harness minimal mode", variant: "max effort", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "CyberGym", score: 76.7, metric: "score", harness: "DeepSeek Harness minimal mode", variant: "max effort", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "DeepSWE", score: 54.4, metric: "resolve rate", harness: "DeepSeek Harness minimal mode", variant: "max effort", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "Toolathlon-Verified", score: 70.3, metric: "score", harness: "DeepSeek Harness minimal mode", variant: "max effort", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "Agents' Last Exam", score: 25.2, metric: "score", harness: "DeepSeek Harness minimal mode", variant: "max effort", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "AutomationBench", score: 25.1, metric: "success rate", variant: "max effort", dataset: "public", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "DSBench-FullStack", score: 68.7, metric: "score", variant: "max effort", dataset: "internal", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "DSBench-Hard", score: 59.6, metric: "score", variant: "max effort", dataset: "internal", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }] }, "deepseek/deepseek-v4-pro-0423": { id: "deepseek/deepseek-v4-pro-0423", name: "DeepSeek V4 Pro 0423", description: "DeepSeek V4 Pro initial snapshot with million-token context and support for thinking and non-thinking modes", family: "deepseek-thinking", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-05", release_date: "2026-04-23", last_updated: "2026-04-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 } }, "deepseek/deepseek-v4-flash-vision-exp": { id: "deepseek/deepseek-v4-flash-vision-exp", name: "DeepSeek V4 Flash Vision Exp", description: "Experimental multimodal DeepSeek V4 Flash model for image understanding, coding, and agentic work", family: "deepseek-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-21", last_updated: "2026-08-21", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 384e3 } }, "arcee-ai/trinity-large-preview": { id: "arcee-ai/trinity-large-preview", name: "Trinity Large Preview", description: "Lightly post-trained 398B MoE chat model for creative work, long-context prompts, and tool-using agents", family: "trinity", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2026-01-27", last_updated: "2026-05-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 524288, output: 262144 }, license: "OpenMDW-1.1", links: [{ label: "Model card", url: "https://huggingface.co/arcee-ai/Trinity-Large-Preview", type: "model_card" }, { label: "Announcement", url: "https://www.arcee.ai/blog/trinity-large", type: "announcement" }, { label: "License", url: "https://huggingface.co/arcee-ai/Trinity-Large-Preview/blob/main/LICENSE", type: "license" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/arcee-ai/Trinity-Large-Preview", format: "safetensors" }] }, "arcee-ai/trinity-mini": { id: "arcee-ai/trinity-mini", name: "Trinity Mini", description: "Reasoning-tuned 26B MoE model with 3B active parameters for agents, tools, and multi-step workloads", family: "trinity", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-12-01", last_updated: "2026-05-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, license: "OpenMDW-1.1", links: [{ label: "Model card", url: "https://huggingface.co/arcee-ai/Trinity-Mini", type: "model_card" }, { label: "Announcement", url: "https://www.arcee.ai/blog/the-trinity-manifesto", type: "announcement" }, { label: "License", url: "https://huggingface.co/arcee-ai/Trinity-Mini/blob/main/LICENSE", type: "license" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/arcee-ai/Trinity-Mini", format: "safetensors" }] }, "arcee-ai/trinity-large-thinking": { id: "arcee-ai/trinity-large-thinking", name: "Trinity Large Thinking", description: "Reasoning-optimized 398B MoE agent model with extended thinking for long-horizon and multi-turn tool use", family: "trinity", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-04-01", last_updated: "2026-05-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 524288, output: 262144 }, license: "OpenMDW-1.1", links: [{ label: "Model card", url: "https://huggingface.co/arcee-ai/Trinity-Large-Thinking", type: "model_card" }, { label: "Announcement", url: "https://www.arcee.ai/blog/trinity-large-thinking", type: "announcement" }, { label: "License", url: "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/blob/main/LICENSE", type: "license" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/arcee-ai/Trinity-Large-Thinking", format: "safetensors" }] }, "arcee-ai/trinity-nano-preview": { id: "arcee-ai/trinity-nano-preview", name: "Trinity Nano Preview", description: "Experimental chat-tuned 6B MoE model with 1B active parameters for low-resource chat and instruction following", family: "trinity", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2025-12-01", last_updated: "2026-05-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, license: "OpenMDW-1.1", links: [{ label: "Model card", url: "https://huggingface.co/arcee-ai/Trinity-Nano-Preview", type: "model_card" }, { label: "Announcement", url: "https://www.arcee.ai/blog/the-trinity-manifesto", type: "announcement" }, { label: "License", url: "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/blob/main/LICENSE", type: "license" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/arcee-ai/Trinity-Nano-Preview", format: "safetensors" }] }, "google/gemini-3.1-flash-tts-preview": { id: "google/gemini-3.1-flash-tts-preview", name: "Gemini 3.1 Flash TTS Preview", description: "Low-latency speech generation with steerable prompts and expressive audio tags", family: "gemini-flash", attachment: false, reasoning: false, tool_call: false, temperature: true, knowledge: "2025-01", release_date: "2026-04-15", last_updated: "2026-04-15", modalities: { input: ["text"], output: ["audio"] }, open_weights: false, limit: { context: 8192, output: 16384 } }, "google/gemma-4-26b-a4b-it": { id: "google/gemma-4-26b-a4b-it", name: "Gemma 4 26B A4B IT", description: "Open Gemma instruction model for efficient chat and self-hosted deployments", family: "gemma", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-02", last_updated: "2026-04-02", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/google/gemma-4-26B-A4B-it" }] }, "google/gemini-3-pro-image": { id: "google/gemini-3-pro-image", name: "Nano Banana Pro", description: "Nano Banana Pro for higher-fidelity image generation and design-heavy edits", family: "gemini-pro", attachment: true, reasoning: true, tool_call: false, temperature: true, knowledge: "2025-01", release_date: "2026-05-28", last_updated: "2026-05-28", modalities: { input: ["text", "image"], output: ["text", "image"] }, open_weights: false, limit: { context: 65536, output: 32768 } }, "google/gemini-3-pro-image-preview": { id: "google/gemini-3-pro-image-preview", name: "Nano Banana Pro", description: "Nano Banana Pro for higher-fidelity image generation and design-heavy edits", family: "gemini-pro", attachment: true, reasoning: true, tool_call: false, temperature: true, knowledge: "2025-01", release_date: "2025-11-20", last_updated: "2025-11-20", modalities: { input: ["text", "image"], output: ["text", "image"] }, open_weights: false, limit: { context: 65536, output: 32768 } }, "google/gemini-flash-lite-latest": { id: "google/gemini-flash-lite-latest", name: "Gemini Flash-Lite Latest", description: "Fast Gemini model balancing multimodal reasoning, tool use, and cost", family: "gemini-flash-lite", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-03", release_date: "2026-07-21", last_updated: "2026-07-21", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemini-2.0-flash": { id: "google/gemini-2.0-flash", name: "Gemini 2.0 Flash", description: "Earlier Gemini Flash workhorse for responsive multimodal apps and tool use", family: "gemini-flash", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-06", release_date: "2024-12-11", last_updated: "2024-12-11", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 8192 } }, "google/gemini-2.5-flash-tts": { id: "google/gemini-2.5-flash-tts", name: "Gemini 2.5 Flash TTS", description: "Speech generation model for controllable voice, narration, and audio delivery", family: "gemini-flash", attachment: false, reasoning: false, tool_call: false, temperature: true, knowledge: "2025-01", release_date: "2025-09-30", last_updated: "2025-12-10", modalities: { input: ["text"], output: ["audio"] }, open_weights: false, limit: { context: 32768, output: 16384 } }, "google/gemini-3.5-flash-lite": { id: "google/gemini-3.5-flash-lite", name: "Gemini 3.5 Flash Lite", description: "Fast Gemini model balancing multimodal reasoning, tool use, and cost", family: "gemini-flash-lite", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-03", release_date: "2026-07-21", last_updated: "2026-07-21", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "SWE-Bench Pro", score: 54.2, metric: "resolve rate", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "Terminal-Bench", score: 54, metric: "accuracy", harness: "Terminus 2", version: "2.1", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "MLE-Bench", score: 39.2, metric: "average position score", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "GDPval-AA", score: 1140, metric: "Elo", version: "v2", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "OSWorld-Verified", score: 74, metric: "success rate", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "CharXiv Reasoning", score: 74.5, metric: "accuracy", variant: "no tools", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "CharXiv Reasoning", score: 76.5, metric: "accuracy", variant: "with tools", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "GDM-MRCR", score: 72.2, metric: "accuracy", variant: "128k average, 8-needle", version: "v2", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "GDM-MRCR", score: 21.3, metric: "accuracy", variant: "1M pointwise, 8-needle", version: "v2", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }] }, "google/gemini-3.1-flash-lite-image": { id: "google/gemini-3.1-flash-lite-image", name: "Nano Banana 2 Lite", description: "Fastest, most cost-efficient Gemini image model for high-volume 1K generation and editing", family: "gemini-flash-lite", attachment: true, reasoning: true, tool_call: true, structured_output: false, temperature: true, knowledge: "2025-01", release_date: "2026-06-30", last_updated: "2026-06-30", modalities: { input: ["text", "image"], output: ["text", "image"] }, open_weights: false, limit: { context: 65536, output: 4096 } }, "google/gemini-3.1-pro-preview": { id: "google/gemini-3.1-pro-preview", name: "Gemini 3.1 Pro Preview", description: "Reasoning-first Gemini preview for agentic coding and complex problem solving", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-02-19", last_updated: "2026-02-19", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "SWE-Bench Pro", score: 54.2, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "Terminal-Bench", score: 70.3, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "SWE-Bench Pro", score: 46.1, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }, { name: "SWE-Atlas Codebase QnA", score: 13.5, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 33.81, metric: "score", harness: "Gemini CLI", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 29.84, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "Artificial Analysis Coding Agent Index", score: 43, metric: "average pass@1", harness: "Gemini CLI", variant: "high", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 45.6, metric: "pass@1", harness: "Gemini CLI", variant: "high", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 15.1, metric: "pass@1", harness: "Gemini CLI", variant: "high", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 68.3, metric: "pass@1", harness: "Gemini CLI", variant: "high", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "GPQA Diamond", score: 94.3, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 44.4, metric: "accuracy", dataset: "full set, text + MM", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "ARC-AGI-2", score: 77.1, metric: "accuracy", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "MMMU Pro", score: 80.5, metric: "accuracy", variant: "no tools", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "MCP Atlas", score: 78.2, metric: "success rate", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "OSWorld-Verified", score: 76.2, metric: "success rate", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "CharXiv Reasoning", score: 83.3, metric: "accuracy", variant: "no tools", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "GDPval-AA", score: 1314, metric: "Elo", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }] }, "google/gemini-3.1-pro-preview-customtools": { id: "google/gemini-3.1-pro-preview-customtools", name: "Gemini 3.1 Pro Preview Custom Tools", description: "Advanced Gemini model for complex reasoning, coding, and multimodal analysis", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-02-19", last_updated: "2026-02-19", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/deep-research-preview-04-2026": { id: "google/deep-research-preview-04-2026", name: "Gemini Deep Research Preview", description: "Agentic model for autonomous multi-step research, synthesis, and cited reports", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2025-01", release_date: "2026-04-21", last_updated: "2026-04-21", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text", "image"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemini-2.5-flash-lite": { id: "google/gemini-2.5-flash-lite", name: "Gemini 2.5 Flash-Lite", description: "Lean Gemini 2.5 lane for cheap multimodal traffic and quick agents", family: "gemini-flash-lite", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2025-06-17", last_updated: "2025-06-17", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 9.5, metric: "index", source: "https://openrouter.ai/google/gemini-2.5-flash-lite/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 19.3, metric: "percent correct", source: "https://openrouter.ai/google/gemini-2.5-flash-lite/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 4.5, metric: "success rate", source: "https://openrouter.ai/google/gemini-2.5-flash-lite/benchmarks", date: "2026-03-11" }] }, "google/gemini-robotics-er-1.6-preview": { id: "google/gemini-robotics-er-1.6-preview", name: "Gemini Robotics-ER 1.6 Preview", description: "Vision-language model for embodied reasoning: spatial understanding, task planning, and physical-world agentic robotics", family: "gemini", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-04-14", last_updated: "2026-04-14", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 65536 } }, "google/gemma-4-31b-it": { id: "google/gemma-4-31b-it", name: "Gemma 4 31B IT", description: "Largest Gemma 4 instruction model for open, self-hosted chat and reasoning", family: "gemma", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-02", last_updated: "2026-04-02", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/google/gemma-4-31B-it" }] }, "google/gemini-2.5-computer-use-preview-10-2025": { id: "google/gemini-2.5-computer-use-preview-10-2025", name: "Gemini 2.5 Computer Use Preview", description: "Specialized Gemini 2.5 model for browser-control agents that automate UI tasks", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-01", release_date: "2025-10-07", last_updated: "2025-10-07", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 64e3 } }, "google/lyria-3-clip-preview": { id: "google/lyria-3-clip-preview", name: "Lyria 3 Clip Preview", description: "Music generation model for short 30-second clips, loops, and previews from text or image prompts", family: "lyria", attachment: true, reasoning: false, tool_call: false, structured_output: false, temperature: true, release_date: "2026-03-25", last_updated: "2026-03-25", modalities: { input: ["text", "image"], output: ["text", "audio"] }, open_weights: false, limit: { context: 131072, output: 65536 } }, "google/veo-3.1-fast-generate-preview": { id: "google/veo-3.1-fast-generate-preview", name: "Veo 3.1 Fast Preview", description: "Video model for prompt-guided generation, editing, and motion workflows", family: "veo", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2025-10-15", last_updated: "2026-01-01", modalities: { input: ["text", "image", "video"], output: ["video"] }, open_weights: false, limit: { context: 1024, output: 0 } }, "google/gemini-embedding-001": { id: "google/gemini-embedding-001", name: "Gemini Embedding 001", description: "Embedding model for semantic search, retrieval, clustering, and ranking pipelines", family: "gemini", attachment: false, reasoning: false, tool_call: false, temperature: false, knowledge: "2025-05", release_date: "2025-05-20", last_updated: "2025-05-20", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 2048, output: 1 } }, "google/gemini-flash-latest": { id: "google/gemini-flash-latest", name: "Gemini Flash Latest", description: "High-efficiency Gemini model for agentic workflows, coding, and multimodal reasoning", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-03", release_date: "2026-08-13", last_updated: "2026-08-13", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemini-3.5-flash": { id: "google/gemini-3.5-flash", name: "Gemini 3.5 Flash", description: "Fast Gemini model balancing multimodal reasoning, tool use, and cost", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-05-19", last_updated: "2026-05-19", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "Terminal-Bench", score: 76.2, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "SWE-Bench Pro", score: 55.1, metric: "resolve rate", variant: "single attempt", dataset: "public", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "MCP Atlas", score: 83.6, metric: "success rate", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "Toolathlon", score: 56.5, metric: "success rate", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "OSWorld-Verified", score: 78.4, metric: "success rate", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "MMMU Pro", score: 83.6, metric: "accuracy", variant: "no tools", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "CharXiv Reasoning", score: 84.2, metric: "accuracy", variant: "no tools", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "Humanity's Last Exam", score: 40.2, metric: "accuracy", dataset: "full set, text + MM", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "ARC-AGI-2", score: 72.1, metric: "accuracy", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "GDPval-AA", score: 1656, metric: "Elo", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }] }, "google/gemini-3.5-live-translate-preview": { id: "google/gemini-3.5-live-translate-preview", name: "Gemini 3.5 Live Translate Preview", description: "Low-latency audio-to-audio model for real-time speech translation across 70+ languages", family: "gemini-pro", attachment: false, reasoning: false, tool_call: false, temperature: false, knowledge: "2025-01", release_date: "2026-06-09", last_updated: "2026-06-09", modalities: { input: ["audio"], output: ["audio", "text"] }, open_weights: false, limit: { context: 131072, output: 65536 } }, "google/veo-3.1-lite-generate-preview": { id: "google/veo-3.1-lite-generate-preview", name: "Veo 3.1 Lite Preview", description: "Video model for prompt-guided generation, editing, and motion workflows", family: "veo", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2026-03-31", last_updated: "2026-03-31", modalities: { input: ["text", "image"], output: ["video"] }, open_weights: false, limit: { context: 1024, output: 0 } }, "google/gemini-omni-flash-preview": { id: "google/gemini-omni-flash-preview", name: "Gemini Omni Flash Preview", description: "Video generation and editing model for fast, conversational text- and image-to-video workflows", family: "gemini", attachment: true, reasoning: true, tool_call: false, release_date: "2026-06-30", last_updated: "2026-06-30", modalities: { input: ["text", "image", "video"], output: ["video"] }, open_weights: false, limit: { context: 1048576, output: 57920 }, benchmarks: [{ name: "LMArena Text-to-Video Arena", score: 1527, metric: "Elo", source: "https://venturebeat.com/technology/googles-gemini-omni-flash-hits-the-api-turning-enterprise-video-production-into-a-conversation", date: "2026-06-30" }] }, "google/gemini-3.1-flash-lite": { id: "google/gemini-3.1-flash-lite", name: "Gemini 3.1 Flash Lite", description: "Low-latency Gemini model for high-volume multimodal and agent workloads", family: "gemini-flash-lite", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-05-07", last_updated: "2026-05-07", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemini-2.5-flash": { id: "google/gemini-2.5-flash", name: "Gemini 2.5 Flash", description: "Fast Gemini workhorse for multimodal apps where latency and price matter", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2025-06-17", last_updated: "2025-06-17", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "Aider Polyglot", score: 55.1, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-25" }, { name: "Artificial Analysis Coding Index", score: 22.2, metric: "index", source: "https://openrouter.ai/google/gemini-2.5-flash/benchmarks", date: "2026-06-02" }, { name: "SciCode", score: 39.4, metric: "percent correct", source: "https://openrouter.ai/google/gemini-2.5-flash/benchmarks", date: "2026-06-02" }, { name: "Terminal-Bench Hard", score: 13.6, metric: "success rate", source: "https://openrouter.ai/google/gemini-2.5-flash/benchmarks", date: "2026-06-02" }] }, "google/gemini-3-pro-preview": { id: "google/gemini-3-pro-preview", name: "Gemini 3 Pro Preview", description: "Preview Gemini flagship for complex reasoning, coding, and rich multimodal prompts", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2025-11-18", last_updated: "2025-11-18", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "SWE-Bench Pro", score: 43.3, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "google/veo-3.1-generate-preview": { id: "google/veo-3.1-generate-preview", name: "Veo 3.1 Preview", description: "Video model for prompt-guided generation, editing, and motion workflows", family: "veo", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2025-10-15", last_updated: "2026-01", modalities: { input: ["text", "image"], output: ["video"] }, open_weights: false, limit: { context: 1024, output: 1 } }, "google/gemini-2.5-pro-tts": { id: "google/gemini-2.5-pro-tts", name: "Gemini 2.5 Pro TTS", description: "Speech generation model for controllable voice, narration, and audio delivery", family: "gemini-pro", attachment: false, reasoning: false, tool_call: false, temperature: false, knowledge: "2025-01", release_date: "2025-09-30", last_updated: "2025-12-10", modalities: { input: ["text"], output: ["audio"] }, open_weights: false, limit: { context: 32768, output: 16384 } }, "google/gemini-embedding-2": { id: "google/gemini-embedding-2", name: "Gemini Embedding 2", description: "Multimodal embedding model mapping text, images, video, audio, and PDFs into a unified embedding space", family: "gemini", attachment: true, reasoning: false, tool_call: false, temperature: false, knowledge: "2025-11", release_date: "2026-04-22", last_updated: "2026-04-22", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 8192, output: 3072 } }, "google/gemma-4-E4B-it": { id: "google/gemma-4-E4B-it", name: "Gemma 4 E4B IT", description: "Open Gemma instruction model for efficient chat and self-hosted deployments", family: "gemma", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-02", last_updated: "2026-04-02", modalities: { input: ["text", "image", "audio"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/google/gemma-4-E4B-it" }] }, "google/gemma-4-E2B-it": { id: "google/gemma-4-E2B-it", name: "Gemma 4 E2B IT", description: "Open Gemma instruction model for efficient chat and self-hosted deployments", family: "gemma", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-02", last_updated: "2026-04-02", modalities: { input: ["text", "image", "audio"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/google/gemma-4-E2B-it" }] }, "google/gemini-3.6-flash": { id: "google/gemini-3.6-flash", name: "Gemini 3.6 Flash", description: "Fast Gemini model balancing multimodal reasoning, tool use, and cost", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-03", release_date: "2026-07-21", last_updated: "2026-07-21", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "SWE-Bench Pro", score: 58.7, metric: "resolve rate", harness: "Antigravity", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "DeepSWE", score: 49, metric: "resolve rate", variant: "high reasoning", version: "1.1", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "Terminal-Bench", score: 78, metric: "accuracy", harness: "Terminus 2", version: "2.1", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "MLE-Bench", score: 63.9, metric: "average position score", dataset: "Partial 30", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "GDPval-AA", score: 1421, metric: "Elo", version: "v2", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "OSWorld-Verified", score: 83, metric: "success rate", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "CharXiv Reasoning", score: 85.2, metric: "accuracy", variant: "no tools", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "CharXiv Reasoning", score: 89.4, metric: "accuracy", variant: "with tools", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "GDM-MRCR", score: 91.8, metric: "accuracy", variant: "128k average, 8-needle", version: "v2", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "GDM-MRCR", score: 54, metric: "accuracy", variant: "1M pointwise, 8-needle", version: "v2", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }] }, "google/gemini-3.1-flash-image": { id: "google/gemini-3.1-flash-image", name: "Nano Banana 2", description: "Image model for prompt-driven generation, editing, and visual design workflows", family: "gemini-flash", attachment: true, reasoning: true, tool_call: false, temperature: true, knowledge: "2025-01", release_date: "2026-05-28", last_updated: "2026-05-28", modalities: { input: ["text", "image", "video", "pdf"], output: ["text", "image"] }, open_weights: false, limit: { context: 131072, output: 32768 } }, "google/gemini-3.1-flash-image-preview": { id: "google/gemini-3.1-flash-image-preview", name: "Nano Banana 2", description: "Image model for prompt-driven generation, editing, and visual design workflows", family: "gemini-flash", attachment: true, reasoning: true, tool_call: false, temperature: true, knowledge: "2025-01", release_date: "2026-02-26", last_updated: "2026-02-26", modalities: { input: ["text", "image", "pdf"], output: ["text", "image"] }, open_weights: false, limit: { context: 65536, output: 65536 } }, "google/gemini-2.0-flash-lite": { id: "google/gemini-2.0-flash-lite", name: "Gemini 2.0 Flash-Lite", description: "Low-latency Gemini model for high-volume multimodal and agent workloads", family: "gemini-flash-lite", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-06", release_date: "2024-12-11", last_updated: "2024-12-11", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 8192 } }, "google/lyria-3-pro-preview": { id: "google/lyria-3-pro-preview", name: "Lyria 3 Pro Preview", description: "Music generation model for full-length songs from text or images with vocals and structure", family: "lyria", attachment: true, reasoning: false, tool_call: false, structured_output: false, temperature: true, release_date: "2026-03-25", last_updated: "2026-03-25", modalities: { input: ["text", "image"], output: ["text", "audio"] }, open_weights: false, limit: { context: 131072, output: 8192 } }, "google/gemini-3.1-flash-lite-preview": { id: "google/gemini-3.1-flash-lite-preview", name: "Gemini 3.1 Flash Lite Preview", description: "Low-latency Gemini model for high-volume multimodal and agent workloads", family: "gemini-flash-lite", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-03-03", last_updated: "2026-03-03", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemini-3-flash-preview": { id: "google/gemini-3-flash-preview", name: "Gemini 3 Flash Preview", description: "New Gemini flash lane bringing frontier-style multimodal reasoning to cheaper runs", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2025-12-17", last_updated: "2025-12-17", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "SWE-Bench Pro", score: 34.63, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }, { name: "SWE-Atlas Codebase QnA", score: 8.2, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 10, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 30.3, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }] }, "google/gemini-2.5-flash-image": { id: "google/gemini-2.5-flash-image", name: "Nano Banana", description: "Nano Banana image model for fast generation, edits, and character-consistent assets", family: "gemini-flash", attachment: true, reasoning: true, tool_call: false, temperature: true, knowledge: "2024-06", release_date: "2025-08-26", last_updated: "2025-08-26", modalities: { input: ["text", "image"], output: ["text", "image"] }, open_weights: false, limit: { context: 32768, output: 32768 } }, "google/gemini-2.5-pro": { id: "google/gemini-2.5-pro", name: "Gemini 2.5 Pro", description: "Google's proven reasoning model for coding, math, and multimodal analysis", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2025-06-17", last_updated: "2025-06-17", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "Aider Polyglot", score: 83.1, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-06-06" }, { name: "Artificial Analysis Coding Index", score: 32, metric: "index", source: "https://openrouter.ai/google/gemini-2.5-pro/benchmarks", date: "2026-06-02" }, { name: "SciCode", score: 42.8, metric: "percent correct", source: "https://openrouter.ai/google/gemini-2.5-pro/benchmarks", date: "2026-06-02" }, { name: "Terminal-Bench Hard", score: 26.5, metric: "success rate", source: "https://openrouter.ai/google/gemini-2.5-pro/benchmarks", date: "2026-06-02" }] }, "google/deep-research-max-preview-04-2026": { id: "google/deep-research-max-preview-04-2026", name: "Deep Research Max Preview", description: "Maximum-comprehensiveness agentic researcher for multi-step investigation, synthesis, and cited reports", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2025-01", release_date: "2026-04-21", last_updated: "2026-04-21", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text", "image"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemini-3.1-flash-live-preview": { id: "google/gemini-3.1-flash-live-preview", name: "Gemini 3.1 Flash Live Preview", description: "High-quality, low-latency Live API model for real-time dialogue and voice-first AI applications", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: false, temperature: true, knowledge: "2025-01", release_date: "2026-03-26", last_updated: "2026-03-26", modalities: { input: ["text", "image", "video", "audio"], output: ["text", "audio"] }, open_weights: false, limit: { context: 131072, output: 65536 } }, "google/gemini-3.7-flash": { id: "google/gemini-3.7-flash", name: "Gemini 3.7 Flash", description: "High-efficiency Gemini model for agentic workflows, coding, and multimodal reasoning", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-03", release_date: "2026-08-13", last_updated: "2026-08-13", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "FrontierCode", score: 43.6, metric: "score", version: "1.1 Main", source: "https://deepmind.google/models/model-cards/gemini-3-7-flash/", date: "2026-08-13" }, { name: "DeepSWE", score: 65.3, metric: "resolve rate", version: "1.1", source: "https://deepmind.google/models/model-cards/gemini-3-7-flash/", date: "2026-08-13" }, { name: "Terminal-Bench", score: 85.8, metric: "accuracy", version: "2.1", source: "https://deepmind.google/models/model-cards/gemini-3-7-flash/", date: "2026-08-13" }, { name: "AutomationBench", score: 30.4, metric: "accuracy", dataset: "private set", source: "https://deepmind.google/models/model-cards/gemini-3-7-flash/", date: "2026-08-13" }, { name: "GDP.pdf", score: 34, metric: "accuracy", source: "https://deepmind.google/models/model-cards/gemini-3-7-flash/", date: "2026-08-13" }, { name: "GDM-MRCR", score: 97, metric: "accuracy", variant: "128k average, 8-needle", version: "v2", source: "https://deepmind.google/models/model-cards/gemini-3-7-flash/", date: "2026-08-13" }] }, "meta/llama-guard-3-8b": { id: "meta/llama-guard-3-8b", name: "Llama-Guard-3-8B", description: "Llama 3.1-based safety classifier for moderating prompts and model responses", family: "llama", attachment: false, reasoning: false, tool_call: false, temperature: true, knowledge: "2023-12", release_date: "2024-07-23", last_updated: "2024-07-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-Guard-3-8B" }] }, "meta/llama-3.3-70b-instruct": { id: "meta/llama-3.3-70b-instruct", name: "Llama-3.3-70B-Instruct", description: "Popular open Llama workhorse for multilingual chat, coding, and self-hosting", family: "llama", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2023-12", release_date: "2024-12-06", last_updated: "2024-12-06", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 10.7, metric: "index", source: "https://openrouter.ai/meta-llama/llama-3.3-70b-instruct/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 26, metric: "percent correct", source: "https://openrouter.ai/meta-llama/llama-3.3-70b-instruct/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 3, metric: "success rate", source: "https://openrouter.ai/meta-llama/llama-3.3-70b-instruct/benchmarks", date: "2026-03-11" }] }, "meta/llama-4-scout-17b-instruct": { id: "meta/llama-4-scout-17b-instruct", name: "Llama 4 Scout 17B Instruct", description: "Open Llama with long-context vision for efficient multimodal agents", family: "llama", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-08", release_date: "2025-04-05", last_updated: "2025-04-05", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 35e5, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct" }] }, "meta/muse-glimmer-30b": { id: "meta/muse-glimmer-30b", name: "Muse Glimmer 30B", description: "Muse Glimmer is a 30-billion-parameter open-weight multimodal model from Meta Superintelligence Labs, distilled from Muse Spark for always-on local agents, tool use, coding, and image understanding.", family: "muse", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-01-04", release_date: "2026-08-10", last_updated: "2026-08-10", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, license: "Apache 2.0", links: [{ label: "Announcement", url: "https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model", type: "announcement" }, { label: "Model card", url: "https://huggingface.co/meta-models/Muse-Glimmer-30B", type: "model_card" }, { label: "Developer docs", url: "https://developer.meta.com/ai/models/muse-glimmer/", type: "docs" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-models/Muse-Glimmer-30B" }], benchmarks: [{ name: "MCP Atlas", score: 75.5, metric: "success rate", variant: "public", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "DeepSearch QA", score: 74.6, source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "SWE-Bench Pro", score: 51.2, metric: "resolve rate", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "SWE-Bench Verified", score: 76, metric: "resolve rate", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "Terminal-Bench", score: 51.7, metric: "success rate", variant: "with terminus2", version: "2.1", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "OSWorld-Verified", score: 65.9, metric: "success rate", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "AIME 2026", score: 94.7, metric: "accuracy", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "GPQA Diamond", score: 83.5, metric: "accuracy", variant: "AA", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "CharXiv Reasoning", score: 78.8, metric: "accuracy", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }] }, "meta/llama-3.1-8b-instruct": { id: "meta/llama-3.1-8b-instruct", name: "Llama-3.1-8B-Instruct", description: "Compact open Llama model for lightweight chat, drafting, and self-hosting", family: "llama", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2023-12", release_date: "2024-07-23", last_updated: "2024-07-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct" }] }, "meta/muse-spark-1.2": { id: "meta/muse-spark-1.2", name: "Muse Spark 1.2", description: "Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1 with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows.", family: "muse", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-05", last_updated: "2026-08-05", modalities: { input: ["text", "image", "video", "pdf", "audio"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 131072 } }, "meta/muse-spark-1.1": { id: "meta/muse-spark-1.1", name: "Muse Spark 1.1", description: "Muse Spark is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration.", family: "muse", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-08", last_updated: "2026-07-09", modalities: { input: ["text", "image", "pdf", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 32e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 61.5, metric: "resolve rate", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "Terminal-Bench", score: 80, metric: "success rate", version: "2.1", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "DeepSWE", score: 53.3, metric: "resolve rate", version: "1.1", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "MCP Atlas", score: 88.1, metric: "success rate", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "JobBench", score: 54.7, metric: "success rate", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "Toolathlon-Verified", score: 75.6, metric: "success rate", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "Humanity's Last Exam", score: 62.1, metric: "accuracy", variant: "with tools", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "OSWorld-Verified", score: 80.8, metric: "success rate", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "Finance Agent", score: 57.2, metric: "accuracy", version: "v2", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "CharXiv Reasoning", score: 88.4, metric: "accuracy", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "BabyVision", score: 76.3, metric: "accuracy", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }] }, "meta/llama-3.2-1b": { id: "meta/llama-3.2-1b", name: "Llama-3.2-1B", description: "Compact open Llama base model for lightweight and on-device use", family: "llama", attachment: false, reasoning: false, tool_call: false, temperature: true, knowledge: "2023-12", release_date: "2024-09-25", last_updated: "2024-09-25", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, license: "Llama 3.2 Community License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-3.2-1B" }] }, "meta/llama-4-maverick-17b-instruct": { id: "meta/llama-4-maverick-17b-instruct", name: "Llama 4 Maverick 17B Instruct", description: "Open multimodal Llama for strong reasoning with efficient everyday serving", family: "llama", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-08", release_date: "2025-04-05", last_updated: "2025-04-05", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct" }], benchmarks: [{ name: "Aider Polyglot", score: 15.6, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-04-06" }, { name: "SWE-Bench Pro", score: 5.24, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "meta/llama-3.2-3b": { id: "meta/llama-3.2-3b", name: "Llama-3.2-3B", description: "Small open Llama base model for lightweight text generation and self-hosting", family: "llama", attachment: false, reasoning: false, tool_call: false, temperature: true, knowledge: "2023-12", release_date: "2024-09-25", last_updated: "2024-09-25", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, license: "Llama 3.2 Community License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-3.2-3B" }] }, "meta/llama-3.2-11b-vision-instruct": { id: "meta/llama-3.2-11b-vision-instruct", name: "Llama-3.2-11B-Vision-Instruct", description: "Open multimodal Llama model for image understanding, captioning, and visual QA", family: "llama", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2023-12", release_date: "2024-09-25", last_updated: "2024-09-25", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct" }] }, "sdaia/allam-2-7b": { id: "sdaia/allam-2-7b", name: "ALLaM-2-7b", description: "ALLaM-2-7b instruction tuned model by SDAIA", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2025-01-23", last_updated: "2025-01-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 4096, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/ALLaM-AI/ALLaM-2.0-7B-Instruct" }] }, "poolside/laguna-m.1": { id: "poolside/laguna-m.1", name: "Laguna M.1", description: "Poolside's open-weight model for agentic coding and long-horizon work", family: "laguna", attachment: false, reasoning: true, tool_call: true, structured_output: false, temperature: true, release_date: "2026-04-28", last_updated: "2026-06-13", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 32768 } }, "poolside/laguna-s-2.1": { id: "poolside/laguna-s-2.1", name: "Laguna S 2.1", description: "Agentic coding model from Poolside in the XS size class for local deployment", family: "laguna", attachment: false, reasoning: true, tool_call: true, structured_output: false, temperature: true, release_date: "2026-07-21", last_updated: "2026-07-21", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 32768 } }, "poolside/laguna-xs-2.1": { id: "poolside/laguna-xs-2.1", name: "Laguna XS 2.1", description: "Agentic coding model from Poolside in the XS size class for local deployment", family: "laguna", attachment: false, reasoning: true, tool_call: true, structured_output: false, temperature: true, release_date: "2026-07-02", last_updated: "2026-07-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 32768 }, benchmarks: [{ name: "SWE-Bench Verified", score: 70.9, metric: "resolved", harness: "Harbor", source: "https://poolside.ai/blog/introducing-laguna-xs-2-1", date: "2026-07-02" }, { name: "SWE-Bench Multilingual", score: 63.1, metric: "resolve rate", harness: "Harbor", source: "https://poolside.ai/blog/introducing-laguna-xs-2-1", date: "2026-07-02" }, { name: "SWE-Bench Pro", score: 47.6, metric: "resolve rate", harness: "Harbor", source: "https://poolside.ai/blog/introducing-laguna-xs-2-1", date: "2026-07-02" }, { name: "Terminal-Bench", score: 37.5, metric: "success rate", harness: "Harbor", version: "2.0", source: "https://poolside.ai/blog/introducing-laguna-xs-2-1", date: "2026-07-02" }] }, "poolside/laguna-xs.2": { id: "poolside/laguna-xs.2", name: "Laguna XS.2", description: "Agentic coding model from Poolside in the XS size class for local deployment", family: "laguna", attachment: false, reasoning: true, tool_call: true, structured_output: false, temperature: true, release_date: "2026-04-28", last_updated: "2026-06-13", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 32768 } }, "bytedance-seed/seed-2.0-code": { id: "bytedance-seed/seed-2.0-code", name: "Seed 2.0 Code", description: "ByteDance Seed coding model for multimodal software engineering and long-running agents", family: "seed", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-14", last_updated: "2026-02-14", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 131072 } }, "bytedance-seed/seed-2.1-turbo": { id: "bytedance-seed/seed-2.1-turbo", name: "Seed 2.1 Turbo", description: "Faster ByteDance Seed 2.1 model for multimodal reasoning and latency-sensitive agent workflows", family: "seed", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-23", last_updated: "2026-06-23", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 256e3 } }, "bytedance-seed/seed-1-6": { id: "bytedance-seed/seed-1-6", name: "Seed 1.6", description: "ByteDance Seed model for long-context reasoning, instruction following, and tool-assisted tasks", family: "seed", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-10-15", last_updated: "2025-10-15", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 64e3 } }, "bytedance-seed/seed-character": { id: "bytedance-seed/seed-character", name: "Seed Character", description: "ByteDance Seed model optimized for character-driven dialogue and consistent conversational behavior", family: "seed", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-23", last_updated: "2026-06-23", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 256e3 } }, "bytedance-seed/seed-1-6-vision": { id: "bytedance-seed/seed-1-6-vision", name: "Seed 1.6 Vision", description: "ByteDance Seed multimodal model for image understanding, visual reasoning, and tool-assisted tasks", family: "seed", attachment: true, reasoning: false, tool_call: true, temperature: true, release_date: "2025-08-15", last_updated: "2025-08-15", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 32e3 } }, "bytedance-seed/seed-2.0-pro": { id: "bytedance-seed/seed-2.0-pro", name: "Seed 2.0 Pro", description: "Flagship ByteDance Seed 2.0 model for complex multimodal reasoning and long-horizon agent workflows", family: "seed", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-02-14", last_updated: "2026-02-14", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 128e3 } }, "bytedance-seed/seed-evolving": { id: "bytedance-seed/seed-evolving", name: "Seed Evolving", description: "Rolling ByteDance Seed model for rapidly updated reasoning, coding, and agent capabilities", family: "seed", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-23", last_updated: "2026-06-23", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 256e3 } }, "bytedance-seed/seed-2.1-pro": { id: "bytedance-seed/seed-2.1-pro", name: "Seed 2.1 Pro", description: "Flagship ByteDance Seed 2.1 model for complex multimodal reasoning, coding, and agents", family: "seed", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-23", last_updated: "2026-06-23", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 256e3 } }, "bytedance-seed/seed-1-6-flash": { id: "bytedance-seed/seed-1-6-flash", name: "Seed 1.6 Flash", description: "Low-latency ByteDance Seed model for high-throughput chat, extraction, and lightweight tool use", family: "seed", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2025-08-28", last_updated: "2025-08-28", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 32e3 } }, "bytedance-seed/seed-2.0-mini": { id: "bytedance-seed/seed-2.0-mini", name: "Seed 2.0 Mini", description: "Lightweight ByteDance Seed 2.0 model for low-latency multimodal reasoning and high-volume tasks", family: "seed", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-02-14", last_updated: "2026-02-14", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 32e3 } }, "bytedance-seed/seed-1-8": { id: "bytedance-seed/seed-1-8", name: "Seed 1.8", description: "ByteDance Seed model for multimodal reasoning, long-context analysis, and agent workflows", family: "seed", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-12-28", last_updated: "2025-12-28", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 64e3 } }, "bytedance-seed/seed-2.0-lite": { id: "bytedance-seed/seed-2.0-lite", name: "Seed 2.0 Lite", description: "Cost-efficient ByteDance Seed 2.0 model for production chat, analysis, and structured generation", family: "seed", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-02-14", last_updated: "2026-02-14", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 32e3 } }, "moonshotai/kimi-k2.7-code": { id: "moonshotai/kimi-k2.7-code", name: "Kimi K2.7 Code", description: "Coding-focused Kimi model, stronger on long-horizon repo work with less overthinking", family: "kimi-k2", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-01", release_date: "2026-06-12", last_updated: "2026-06-12", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/moonshotai/Kimi-K2.7-Code" }], benchmarks: [{ name: "Kimi Code Bench", score: 62, harness: "Kimi Code CLI", version: "v2", source: "https://huggingface.co/moonshotai/Kimi-K2.7-Code", date: "2026-06-12" }, { name: "Program Bench", score: 53.6, harness: "Kimi Code CLI", source: "https://huggingface.co/moonshotai/Kimi-K2.7-Code", date: "2026-06-12" }, { name: "MLS Bench Lite", score: 35.1, harness: "Kimi Code CLI", source: "https://huggingface.co/moonshotai/Kimi-K2.7-Code", date: "2026-06-12" }, { name: "MCP Atlas", score: 76, metric: "success rate", harness: "Kimi Code CLI", source: "https://huggingface.co/moonshotai/Kimi-K2.7-Code", date: "2026-06-12" }, { name: "MCP Mark Verified", score: 81.1, metric: "success rate", harness: "Kimi Code CLI", source: "https://huggingface.co/moonshotai/Kimi-K2.7-Code", date: "2026-06-12" }, { name: "Kimi Claw 24/7 Bench", score: 46.9, harness: "Kimi Code CLI", source: "https://huggingface.co/moonshotai/Kimi-K2.7-Code", date: "2026-06-12" }] }, "moonshotai/kimi-k3": { id: "moonshotai/kimi-k3", name: "Kimi K3", description: "Multimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent work", family: "kimi-k3", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, release_date: "2026-07-16", last_updated: "2026-07-16", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 131072 }, benchmarks: [{ name: "DeepSWE", score: 67.5, metric: "resolve rate", harness: "Kimi Code", variant: "max effort", version: "1.1", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "Terminal-Bench", score: 88.3, metric: "accuracy", harness: "Kimi Code", variant: "max effort", version: "2.1", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "FrontierSWE", score: 81.2, metric: "dominance score", harness: "Kimi Code", variant: "max effort", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "Program Bench", score: 77.8, metric: "score", harness: "Kimi Code", variant: "max effort", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "SWE Marathon", score: 42, metric: "resolve rate", harness: "Claude Code", variant: "max effort", version: "1.1", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "GDPval-AA", score: 1668, metric: "Elo", variant: "max effort", version: "v2", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "AA-Briefcase", score: 1548, metric: "Elo", variant: "max effort", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "AutomationBench", score: 30.8, metric: "success rate", variant: "max effort", dataset: "600-task public subset", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "JobBench", score: 52.9, metric: "score", variant: "max effort", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "SpreadsheetBench", score: 34.8, metric: "score", harness: "Claude Code", variant: "max effort", version: "2", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "BrowseComp", score: 91.2, metric: "accuracy", variant: "max effort, context compaction", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "CharXiv Reasoning", score: 91.3, metric: "accuracy", variant: "max effort, with tools", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "ZeroBench", score: 41, metric: "pass@5", variant: "max effort, with tools", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }] }, "moonshotai/kimi-k2-thinking-turbo": { id: "moonshotai/kimi-k2-thinking-turbo", name: "Kimi K2 Thinking Turbo", description: "Kimi reasoning model for long-horizon research, planning, and tool use", family: "kimi-thinking", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-08", release_date: "2025-11-06", last_updated: "2025-11-06", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/moonshotai/Kimi-K2-Thinking" }] }, "moonshotai/kimi-k2.5": { id: "moonshotai/kimi-k2.5", name: "Kimi K2.5", description: "Earlier Kimi frontier model for long-context agents, coding, and multimodal work", family: "kimi-k2", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-01", release_date: "2026-01", last_updated: "2026-01", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/moonshotai/Kimi-K2.5" }], benchmarks: [{ name: "SWE-Bench Verified", score: 70.8, metric: "resolved", source: "https://www.swebench.com/" }, { name: "SWE-Atlas Codebase QnA", score: 13.1, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 20.95, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 25.77, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }] }, "moonshotai/kimi-k2.7-code-highspeed": { id: "moonshotai/kimi-k2.7-code-highspeed", name: "Kimi K2.7 Code Highspeed", description: "Lower-latency Kimi Code variant for interactive edits and coding-agent loops", family: "kimi-k2", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-01", release_date: "2026-06-12", last_updated: "2026-06-12", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/moonshotai/Kimi-K2.7-Code" }] }, "moonshotai/kimi-k2-thinking": { id: "moonshotai/kimi-k2-thinking", name: "Kimi K2 Thinking", description: "Thinking Kimi model for slower research passes, planning, and hard technical questions", family: "kimi-thinking", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-08", release_date: "2025-11-06", last_updated: "2025-11-06", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/moonshotai/Kimi-K2-Thinking" }], benchmarks: [{ name: "SWE-Bench Verified", score: 71.3, metric: "resolved", source: "https://huggingface.co/moonshotai/Kimi-K2-Thinking" }] }, "moonshotai/kimi-k2.6": { id: "moonshotai/kimi-k2.6", name: "Kimi K2.6", description: "Multimodal Kimi workhorse for agent loops, coding tasks, and visual context", family: "kimi-k2", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-04-21", last_updated: "2026-04-21", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/moonshotai/Kimi-K2.6" }], benchmarks: [{ name: "SWE-Bench Verified", score: 80.2, metric: "resolved", source: "https://huggingface.co/moonshotai/Kimi-K2.6" }, { name: "Artificial Analysis Coding Agent Index", score: 50.5, metric: "average pass@1", harness: "Claude Code", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 59.8, metric: "pass@1", harness: "Claude Code", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 27.3, metric: "pass@1", harness: "Claude Code", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 64.3, metric: "pass@1", harness: "Claude Code", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }] }, "anthropic/claude-opus-4-20250514": { id: "anthropic/claude-opus-4-20250514", name: "Claude Opus 4", description: "Flagship Claude model for deep reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-05-22", last_updated: "2025-05-22", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 32e3 }, benchmarks: [{ name: "Aider Polyglot", score: 72, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-25" }] }, "anthropic/claude-opus-4-7": { id: "anthropic/claude-opus-4-7", name: "Claude Opus 4.7", description: "Stronger Opus tier for advanced software work and high-stakes reasoning", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2026-01-31", release_date: "2026-04-16", last_updated: "2026-04-16", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 64.3, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "Terminal-Bench", score: 66.1, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "SWE-Atlas Refactoring", score: 48.57, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "Artificial Analysis Coding Agent Index", score: 66.6, metric: "average pass@1", harness: "Claude Code", variant: "max", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 81, metric: "pass@1", harness: "Claude Code", variant: "max", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 44.9, metric: "pass@1", harness: "Claude Code", variant: "max", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 73.8, metric: "pass@1", harness: "Claude Code", variant: "max", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Artificial Analysis Coding Agent Index", score: 61.2, metric: "average pass@1", harness: "Cursor CLI", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 78.4, metric: "pass@1", harness: "Cursor CLI", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 34.4, metric: "pass@1", harness: "Cursor CLI", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 70.6, metric: "pass@1", harness: "Cursor CLI", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Artificial Analysis Coding Agent Index", score: 59.9, metric: "average pass@1", harness: "Claude Code", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 71.7, metric: "pass@1", harness: "Claude Code", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 36.4, metric: "pass@1", harness: "Claude Code", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 71.4, metric: "pass@1", harness: "Claude Code", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "GPQA Diamond", score: 94.2, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 46.9, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 54.7, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "OSWorld-Verified", score: 78, metric: "success rate", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }] }, "anthropic/claude-mythos-5": { id: "anthropic/claude-mythos-5", name: "Claude Mythos 5", description: "Restricted Claude model for advanced cybersecurity and biology research workflows", family: "claude-mythos", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2026-01-31", release_date: "2026-06-09", last_updated: "2026-06-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 } }, "anthropic/claude-opus-4-1-20250805": { id: "anthropic/claude-opus-4-1-20250805", name: "Claude Opus 4.1", description: "Flagship Claude model for deep reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-08-05", last_updated: "2025-08-05", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 32e3 } }, "anthropic/claude-3-5-sonnet-20241022": { id: "anthropic/claude-3-5-sonnet-20241022", name: "Claude Sonnet 3.5 v2", description: "Balanced Claude model for coding, analysis, agent workflows, and cost control", family: "claude-sonnet", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-04-30", release_date: "2024-10-22", last_updated: "2024-10-22", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 8192 }, benchmarks: [{ name: "Aider Polyglot", score: 51.6, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-01-17" }] }, "anthropic/claude-opus-4-8": { id: "anthropic/claude-opus-4-8", name: "Claude Opus 4.8", description: "Top Claude Opus tier for the hardest reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2026-01", release_date: "2026-05-28", last_updated: "2026-05-28", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 69.2, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "Terminal-Bench", score: 74.6, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "SWE-Bench Verified", score: 88.6, metric: "resolved", source: "https://benchlm.ai/benchmarks/sweVerified" }, { name: "Humanity's Last Exam", score: 49.8, metric: "accuracy", variant: "no tools", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "Humanity's Last Exam", score: 57.9, metric: "accuracy", variant: "with tools", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "OSWorld-Verified", score: 83.4, metric: "success rate", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "FrontierCode", score: 13.4, metric: "pass rate", variant: "high effort", dataset: "Diamond", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }] }, "anthropic/claude-3-5-haiku-20241022": { id: "anthropic/claude-3-5-haiku-20241022", name: "Claude Haiku 3.5", description: "Fast Claude model for responsive assistance, classification, and lightweight agents", family: "claude-haiku", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-07-31", release_date: "2024-10-22", last_updated: "2024-10-22", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 8192 }, benchmarks: [{ name: "Aider Polyglot", score: 28, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2024-12-21" }] }, "anthropic/claude-opus-4-1": { id: "anthropic/claude-opus-4-1", name: "Claude Opus 4.1 (latest)", description: "Flagship Claude model for deep reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-08-05", last_updated: "2025-08-05", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 32e3 } }, "anthropic/claude-sonnet-5": { id: "anthropic/claude-sonnet-5", name: "Claude Sonnet 5", description: "Everyday Claude agent model for coding, planning, browsing, and general work", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2026-01-31", release_date: "2026-06-30", last_updated: "2026-06-30", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Verified", score: 85.2, metric: "resolved", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "SWE-Bench Pro", score: 63.2, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "SWE-Bench Multilingual", score: 78.3, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "Terminal-Bench", score: 80.4, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "OSWorld-Verified", score: 81.2, metric: "success rate", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "BrowseComp", score: 84.7, metric: "accuracy", variant: "single agent", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "FrontierCode", score: 38.8, metric: "pass rate", version: "v1", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }] }, "anthropic/claude-3-7-sonnet-20250219": { id: "anthropic/claude-3-7-sonnet-20250219", name: "Claude Sonnet 3.7", description: "Balanced Claude model for coding, analysis, agent workflows, and cost control", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-10-31", release_date: "2025-02-19", last_updated: "2025-02-19", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 }, benchmarks: [{ name: "Aider Polyglot", score: 64.9, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-02-24" }] }, "anthropic/claude-sonnet-4-5-20250929": { id: "anthropic/claude-sonnet-4-5-20250929", name: "Claude Sonnet 4.5", description: "Balanced Claude model for coding, analysis, agent workflows, and cost control", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-07-31", release_date: "2025-09-29", last_updated: "2025-09-29", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 } }, "anthropic/claude-opus-4-5": { id: "anthropic/claude-opus-4-5", name: "Claude Opus 4.5 (latest)", description: "Flagship Claude model for deep reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-05", release_date: "2025-11-24", last_updated: "2025-11-24", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 } }, "anthropic/claude-sonnet-4-6": { id: "anthropic/claude-sonnet-4-6", name: "Claude Sonnet 4.6", description: "Claude workhorse for coding agents, careful analysis, and production cost control", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-08-31", release_date: "2026-02-17", last_updated: "2026-03-13", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 64e3 }, benchmarks: [{ name: "SWE-Atlas Codebase QnA", score: 31.2, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 32.21, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 31.76, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "Artificial Analysis Coding Agent Index", score: 49.4, metric: "average pass@1", harness: "Claude Code", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 70.3, metric: "pass@1", harness: "Claude Code", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 14.9, metric: "pass@1", harness: "Claude Code", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 63.1, metric: "pass@1", harness: "Claude Code", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 67, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "Humanity's Last Exam", score: 34.6, metric: "accuracy", variant: "no tools", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "Humanity's Last Exam", score: 46.8, metric: "accuracy", variant: "with tools", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "OSWorld-Verified", score: 78.5, metric: "success rate", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }] }, "anthropic/claude-opus-4-0": { id: "anthropic/claude-opus-4-0", name: "Claude Opus 4 (latest)", description: "Flagship Claude model for deep reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-05-22", last_updated: "2025-05-22", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 32e3 }, benchmarks: [{ name: "Aider Polyglot", score: 72, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-25" }] }, "anthropic/claude-haiku-4-5-20251001": { id: "anthropic/claude-haiku-4-5-20251001", name: "Claude Haiku 4.5", description: "Fast Claude model for responsive assistance, classification, and lightweight agents", family: "claude-haiku", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-02-28", release_date: "2025-10-15", last_updated: "2025-10-15", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 } }, "anthropic/claude-sonnet-4-5": { id: "anthropic/claude-sonnet-4-5", name: "Claude Sonnet 4.5 (latest)", description: "Balanced Claude model for coding, analysis, agent workflows, and cost control", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-07-31", release_date: "2025-09-29", last_updated: "2025-09-29", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 43.6, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "anthropic/claude-opus-4-5-20251101": { id: "anthropic/claude-opus-4-5-20251101", name: "Claude Opus 4.5", description: "Flagship Claude model for deep reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-05", release_date: "2025-11-01", last_updated: "2025-11-01", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 45.89, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "anthropic/claude-fable-5": { id: "anthropic/claude-fable-5", name: "Claude Fable 5", description: "Claude model for creative writing, analysis, and controlled agent workflows", family: "claude-fable", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2026-01-31", release_date: "2026-06-09", last_updated: "2026-06-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 80.3, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "SWE-Bench Verified", score: 95, metric: "resolved", source: "https://benchlm.ai/benchmarks/sweVerified" }, { name: "Terminal-Bench", score: 88, metric: "success rate", version: "2.1", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "Humanity's Last Exam", score: 59, metric: "accuracy", variant: "no tools", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "Humanity's Last Exam", score: 64.5, metric: "accuracy", variant: "with tools", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "OSWorld-Verified", score: 85, metric: "success rate", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "FrontierCode", score: 29.3, metric: "pass rate", variant: "high effort", dataset: "Diamond", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "GDPval-AA", score: 1932, metric: "Elo", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "AutomationBench", score: 17.4, metric: "success rate", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }] }, "anthropic/claude-sonnet-4-20250514": { id: "anthropic/claude-sonnet-4-20250514", name: "Claude Sonnet 4", description: "Balanced Claude model for coding, analysis, agent workflows, and cost control", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-05-22", last_updated: "2025-05-22", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 }, benchmarks: [{ name: "Aider Polyglot", score: 61.3, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-24" }] }, "anthropic/claude-sonnet-4-0": { id: "anthropic/claude-sonnet-4-0", name: "Claude Sonnet 4 (latest)", description: "Balanced Claude model for coding, analysis, agent workflows, and cost control", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-05-22", last_updated: "2025-05-22", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 }, benchmarks: [{ name: "Aider Polyglot", score: 61.3, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-24" }, { name: "SWE-Bench Pro", score: 42.7, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "anthropic/claude-haiku-4-5": { id: "anthropic/claude-haiku-4-5", name: "Claude Haiku 4.5 (latest)", description: "Fast Claude lane for lightweight agents, office tasks, and responsive chat", family: "claude-haiku", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-02-28", release_date: "2025-10-15", last_updated: "2025-10-15", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 39.45, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "anthropic/claude-opus-4-6": { id: "anthropic/claude-opus-4-6", name: "Claude Opus 4.6", description: "High-end Claude for difficult coding, planning, and slower expert reasoning", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-05-31", release_date: "2026-02-05", last_updated: "2026-03-13", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 51.9, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }, { name: "SWE-Atlas Codebase QnA", score: 33.3, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Codebase QnA", score: 30, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 35.58, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 36.67, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "SWE-Atlas Test Writing", score: 36.08, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "Artificial Analysis Coding Agent Index", score: 51.3, metric: "average pass@1", harness: "Claude Code", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 71.9, metric: "pass@1", harness: "Claude Code", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 11.8, metric: "pass@1", harness: "Claude Code", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 70.2, metric: "pass@1", harness: "Claude Code", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }] }, "anthropic/claude-3-haiku-20240307": { id: "anthropic/claude-3-haiku-20240307", name: "Claude Haiku 3", description: "Legacy model retained for compatibility with older integrations", family: "claude-haiku", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2023-08-31", release_date: "2024-03-13", last_updated: "2024-03-13", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 4096 } }, "anthropic/claude-opus-5": { id: "anthropic/claude-opus-5", name: "Claude Opus 5", description: "Strongest Claude Opus model for coding, agents, and professional work", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2026-05", release_date: "2026-07-24", last_updated: "2026-07-24", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Verified", score: 96, metric: "resolved", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "SWE-Bench Pro", score: 79.2, metric: "resolve rate", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "SWE-Bench Multilingual", score: 89.5, metric: "resolve rate", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "SWE-Bench Multimodal", score: 59.4, metric: "resolve rate", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "DeepSWE", score: 68.8, metric: "resolve rate", version: "1.1", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "FrontierCode", score: 53.4, metric: "mean@5", variant: "medium effort", dataset: "Main", version: "1.1", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "Frontier-Bench", score: 43.3, metric: "mean reward", harness: "mini-SWE-agent", variant: "max effort", version: "v0.1", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "BrowseComp", score: 90.8, metric: "accuracy", variant: "single agent", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "Humanity's Last Exam", score: 56.3, metric: "accuracy", variant: "no tools", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "Humanity's Last Exam", score: 64.7, metric: "accuracy", variant: "with tools", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "DeepSearchQA", score: 95, metric: "F1", variant: "max effort", source: "https://www.anthropic.com/news/claude-opus-5", date: "2026-07-24" }, { name: "OSWorld", score: 70.6, metric: "success rate", version: "2.0", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "GDPval-AA", score: 1861, metric: "Elo", variant: "max effort", version: "v2", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "AA-Briefcase", score: 1720, metric: "Elo", variant: "max effort", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "AutomationBench", score: 26, metric: "success rate", variant: "max effort", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "ARC-AGI-1", score: 97.5, metric: "accuracy", variant: "max effort", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "ARC-AGI-2", score: 90.4, metric: "accuracy", variant: "max effort", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "ARC-AGI-3", score: 30.2, metric: "RHAE", variant: "high effort", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "HealthBench Professional", score: 59.8, metric: "score", variant: "max effort", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }] }, "cohere/c4ai-aya-expanse-32b": { id: "cohere/c4ai-aya-expanse-32b", name: "Aya Expanse 32B", description: "Open multilingual model optimized for generation across 23 languages", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2024-10-24", last_updated: "2024-10-24", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4e3 }, license: "CC-BY-NC-4.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/aya-expanse-32b" }] }, "cohere/command-a-reasoning-08-2025": { id: "cohere/command-a-reasoning-08-2025", name: "Command A Reasoning", description: "Cohere reasoning model for multilingual enterprise agents, tools, and complex workflows", family: "command-a", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2025-08-21", last_updated: "2025-08-21", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 32e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-a-reasoning-08-2025" }] }, "cohere/c4ai-aya-vision-32b": { id: "cohere/c4ai-aya-vision-32b", name: "Aya Vision 32B", description: "Open multilingual vision model for OCR, visual reasoning, and image question answering", attachment: true, reasoning: false, tool_call: false, temperature: true, release_date: "2025-03-04", last_updated: "2025-05-14", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 16e3, output: 4e3 }, license: "CC-BY-NC-4.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/aya-vision-32b" }] }, "cohere/command-r-plus-08-2024": { id: "cohere/command-r-plus-08-2024", name: "Command R+", description: "Cohere's RAG workhorse for long-context enterprise search and tool use", family: "command-r", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2024-08-30", last_updated: "2024-08-30", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-r-plus-08-2024" }] }, "cohere/command-a-translate-08-2025": { id: "cohere/command-a-translate-08-2025", name: "Command A Translate", description: "Translation model for multilingual conversion, localization, and cross-language workflows", family: "command-a", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2025-08-28", last_updated: "2025-08-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 8e3, output: 8e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-a-translate-08-2025" }] }, "cohere/command-r7b-arabic-02-2025": { id: "cohere/command-r7b-arabic-02-2025", name: "Command R7B Arabic", description: "Open Command R model optimized for Arabic enterprise chat, RAG, and cultural knowledge", family: "command-r", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2025-02-27", last_updated: "2025-02-27", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-r7b-arabic-02-2025" }] }, "cohere/c4ai-aya-expanse-8b": { id: "cohere/c4ai-aya-expanse-8b", name: "Aya Expanse 8B", description: "Compact open multilingual model optimized for generation across 23 languages", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2024-10-24", last_updated: "2024-10-24", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 8e3, output: 4e3 }, license: "CC-BY-NC-4.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/aya-expanse-8b" }] }, "cohere/command-a-vision-07-2025": { id: "cohere/command-a-vision-07-2025", name: "Command A Vision", description: "Cohere vision model for multilingual document analysis, OCR, and image understanding", family: "command-a", attachment: true, reasoning: false, tool_call: false, temperature: true, knowledge: "2024-06-01", release_date: "2025-07-31", last_updated: "2025-07-31", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-a-vision-07-2025" }] }, "cohere/command-a-plus-05-2026": { id: "cohere/command-a-plus-05-2026", name: "Command A Plus", description: "Cohere's stronger command model for multilingual agents and enterprise workflows", family: "command-a", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-04-01", release_date: "2026-05-20", last_updated: "2026-06-09", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 64e3 } }, "cohere/command-a-03-2025": { id: "cohere/command-a-03-2025", name: "Command A", description: "Cohere command model for multilingual enterprise agents, tools, and chat", family: "command-a", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2025-03-13", last_updated: "2025-03-13", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 8e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-a-03-2025" }], benchmarks: [{ name: "Aider Polyglot", score: 12, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-03-14" }] }, "cohere/command-r7b-12-2024": { id: "cohere/command-r7b-12-2024", name: "Command R7B", description: "Cohere retrieval model for long-context chat and enterprise RAG workflows", family: "command-r", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2024-12-02", last_updated: "2024-12-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-r7b-12-2024" }] }, "cohere/command-r-08-2024": { id: "cohere/command-r-08-2024", name: "Command R", description: "Cohere retrieval model for long-context chat and enterprise RAG workflows", family: "command-r", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2024-08-30", last_updated: "2024-08-30", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-r-08-2024" }] }, "cohere/north-mini-code-1-0": { id: "cohere/north-mini-code-1-0", name: "North Mini Code", description: "Cohere coding model for practical software engineering and agentic edits", family: "north", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-09-23", release_date: "2026-06-09", last_updated: "2026-06-09", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 64e3 }, benchmarks: [{ name: "SWE-Bench Verified", score: 67.6, metric: "resolved", harness: "SWE-agent", source: "https://huggingface.co/CohereLabs/North-Mini-Code-1.0", date: "2026-06-09" }, { name: "SWE-Bench Pro", score: 40.2, metric: "resolve rate", harness: "SWE-agent", source: "https://huggingface.co/CohereLabs/North-Mini-Code-1.0", date: "2026-06-09" }, { name: "Artificial Analysis Intelligence Index", score: 27.6, metric: "index score", source: "https://artificialanalysis.ai/articles/north-mini-code-cohere-s-small-coding-focused-moe-model", date: "2026-06-09" }, { name: "Artificial Analysis Coding Index", score: 33.4, metric: "index score", source: "https://artificialanalysis.ai/articles/north-mini-code-cohere-s-small-coding-focused-moe-model", date: "2026-06-09" }, { name: "GDPval-AA", score: 14, metric: "win rate", source: "https://artificialanalysis.ai/articles/north-mini-code-cohere-s-small-coding-focused-moe-model", date: "2026-06-09" }, { name: "\u03C4\xB2-Bench Telecom", score: 37, metric: "success rate", source: "https://artificialanalysis.ai/articles/north-mini-code-cohere-s-small-coding-focused-moe-model", date: "2026-06-09" }] }, "cohere/c4ai-aya-vision-8b": { id: "cohere/c4ai-aya-vision-8b", name: "Aya Vision 8B", description: "Compact open multilingual vision model for OCR and visual question answering", attachment: true, reasoning: false, tool_call: false, temperature: true, release_date: "2025-03-04", last_updated: "2025-05-14", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 16e3, output: 4e3 }, license: "CC-BY-NC-4.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/aya-vision-8b" }] }, "openai/gpt-4o": { id: "openai/gpt-4o", name: "GPT-4o", description: "Omni-era GPT for multimodal chat, practical coding, and general assistants", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2023-09", release_date: "2024-05-13", last_updated: "2024-08-06", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 }, benchmarks: [{ name: "Aider Polyglot", score: 23.1, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2024-12-30" }] }, "openai/gpt-image-1.5": { id: "openai/gpt-image-1.5", name: "GPT-Image-1.5", description: "Image model for prompt-driven generation, editing, and visual design workflows", family: "gpt-image", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2025-11-25", last_updated: "2025-11-25", modalities: { input: ["text", "image"], output: ["text", "image"] }, open_weights: false, limit: { context: 0, output: 0 } }, "openai/gpt-5.3-chat-latest": { id: "openai/gpt-5.3-chat-latest", name: "GPT-5.3 Chat (latest)", description: "Chat-tuned GPT model for conversational assistance, writing, and tool workflows", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-08-31", release_date: "2026-03-03", last_updated: "2026-03-03", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 } }, "openai/gpt-5-nano": { id: "openai/gpt-5-nano", name: "GPT-5 Nano", description: "Tiny GPT-5 lane for routing, extraction, classification, and bulk jobs", family: "gpt-nano", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-05-30", release_date: "2025-08-07", last_updated: "2025-08-07", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-4o-mini": { id: "openai/gpt-4o-mini", name: "GPT-4o mini", description: "Small omni GPT for cheap multimodal assistance and production-scale traffic", family: "gpt-mini", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2023-09", release_date: "2024-07-18", last_updated: "2024-07-18", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 }, benchmarks: [{ name: "Aider Polyglot", score: 3.6, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2024-12-21" }, { name: "SciCode", score: 22.9, metric: "percent correct", source: "https://openrouter.ai/openai/gpt-4o-mini/benchmarks", date: "2026-03-11" }] }, "openai/gpt-3.5-turbo": { id: "openai/gpt-3.5-turbo", name: "GPT-3.5-turbo", description: "Compact GPT model for low-latency assistance and high-volume workloads", family: "gpt", attachment: false, reasoning: false, tool_call: false, structured_output: false, temperature: true, knowledge: "2021-09-01", release_date: "2023-03-01", last_updated: "2023-11-06", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 16385, output: 4096 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 10.7, metric: "index", source: "https://openrouter.ai/openai/gpt-3.5-turbo/benchmarks", date: "2026-03-11" }] }, "openai/o1-pro": { id: "openai/o1-pro", name: "o1-pro", description: "O-series reasoning model for hard analysis, math, coding, and planning", family: "o-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2023-09", release_date: "2025-03-19", last_updated: "2025-03-19", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 } }, "openai/whisper-large-v3": { id: "openai/whisper-large-v3", name: "Whisper 3 Large", description: "Open Whisper checkpoint for robust multilingual transcription and captioning", family: "whisper", attachment: false, reasoning: false, tool_call: false, release_date: "2024-10-01", last_updated: "2024-10-01", modalities: { input: ["audio"], output: ["text"] }, open_weights: true, limit: { context: 448, output: 4096 } }, "openai/gpt-5.5-pro": { id: "openai/gpt-5.5-pro", name: "GPT-5.5 Pro", description: "Highest-accuracy GPT-5.5 tier for slower, precision-heavy reasoning and coding", family: "gpt-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-12-01", release_date: "2026-04-23", last_updated: "2026-04-23", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "BrowseComp", score: 90.1, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 43.1, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 57.2, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 52.4, metric: "accuracy", dataset: "Tier 1-3", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 39.6, metric: "accuracy", dataset: "Tier 4", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GDPval", score: 82.3, metric: "wins or ties", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GeneBench", score: 33.2, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }] }, "openai/gpt-4-turbo": { id: "openai/gpt-4-turbo", name: "GPT-4 Turbo", description: "Compact GPT model for low-latency assistance and high-volume workloads", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: false, temperature: true, knowledge: "2023-12", release_date: "2023-11-06", last_updated: "2024-04-09", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 4096 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 21.5, metric: "index", source: "https://openrouter.ai/openai/gpt-4-turbo/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 31.9, metric: "percent correct", source: "https://openrouter.ai/openai/gpt-4-turbo/benchmarks", date: "2026-03-11" }] }, "openai/gpt-4": { id: "openai/gpt-4", name: "GPT-4", description: "GPT model for general reasoning, writing, coding, and tool-assisted tasks", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: false, temperature: true, knowledge: "2023-11", release_date: "2023-11-06", last_updated: "2024-04-09", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 8192, output: 8192 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 13.1, metric: "index", source: "https://openrouter.ai/openai/gpt-4/benchmarks", date: "2026-03-11" }] }, "openai/gpt-5-chat-latest": { id: "openai/gpt-5-chat-latest", name: "GPT-5 Chat (latest)", description: "Chat-tuned GPT model for conversational assistance, writing, and tool workflows", family: "gpt-codex", attachment: true, reasoning: true, tool_call: false, structured_output: true, temperature: true, knowledge: "2024-09-30", release_date: "2025-08-07", last_updated: "2025-08-07", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-5.6-sol": { id: "openai/gpt-5.6-sol", name: "GPT-5.6 Sol", description: "Frontier GPT-5.6 model for complex professional work, coding, and agentic workflows", family: "gpt-sol", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2026-02-16", release_date: "2026-07-09", last_updated: "2026-07-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 64.6, metric: "resolve rate", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Terminal-Bench", score: 88.8, metric: "success rate", version: "2.1", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "DeepSWE", score: 72.7, metric: "resolve rate", version: "1.1", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "GPQA Diamond", score: 94.6, metric: "accuracy", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "FrontierMath", score: 89, metric: "accuracy", dataset: "Tier 1-3", version: "v2", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "BrowseComp", score: 90.4, metric: "accuracy", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "OSWorld", score: 62.6, metric: "success rate", version: "2.0", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "MMMU Pro", score: 83, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Agents' Last Exam", score: 52.7, source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Toolathlon", score: 58, metric: "success rate", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Artificial Analysis Intelligence Index", score: 58.9, metric: "index score", variant: "max", version: "4.1", source: "https://artificialanalysis.ai/articles/gpt-5-6-has-landed", date: "2026-07-09" }, { name: "Artificial Analysis Coding Agent Index", score: 80, metric: "index score", harness: "Codex", variant: "max", version: "1.1", source: "https://artificialanalysis.ai/articles/gpt-5-6-has-landed", date: "2026-07-09" }] }, "openai/o3-pro": { id: "openai/o3-pro", name: "o3-pro", description: "High-effort o3 tier for difficult technical reasoning and careful answers", family: "o-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-05", release_date: "2025-06-10", last_updated: "2025-06-10", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 }, benchmarks: [{ name: "Aider Polyglot", score: 84.9, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-06-28" }] }, "openai/o3-deep-research": { id: "openai/o3-deep-research", name: "o3-deep-research", description: "Research model for long-horizon investigation, synthesis, and analytical reports", family: "o", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2024-05", release_date: "2024-06-26", last_updated: "2024-06-26", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 } }, "openai/gpt-5.4-pro": { id: "openai/gpt-5.4-pro", name: "GPT-5.4 Pro", description: "More exact GPT-5.4 tier for demanding professional reasoning and agent tasks", family: "gpt-pro", attachment: true, reasoning: true, tool_call: true, structured_output: false, temperature: false, knowledge: "2025-08-31", release_date: "2026-03-05", last_updated: "2026-03-05", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "GPQA Diamond", score: 94.4, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 42.7, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 58.7, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "BrowseComp", score: 89.3, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GDPval", score: 82, metric: "wins or ties", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 50, metric: "accuracy", dataset: "Tier 1-3", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 38, metric: "accuracy", dataset: "Tier 4", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "ARC-AGI-1", score: 94.5, metric: "accuracy", variant: "Verified", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "ARC-AGI-2", score: 83.3, metric: "accuracy", variant: "Verified", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FinanceAgent", score: 61.5, metric: "accuracy", version: "1.1", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GeneBench", score: 25.6, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }] }, "openai/gpt-5": { id: "openai/gpt-5", name: "GPT-5", description: "Original GPT-5 workhorse for reasoning, coding, writing, and tool workflows", family: "gpt", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-08-07", last_updated: "2025-08-07", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "Aider Polyglot", score: 88, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-08-23" }, { name: "SWE-Bench Pro", score: 41.78, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "openai/gpt-5.1-codex-mini": { id: "openai/gpt-5.1-codex-mini", name: "GPT-5.1 Codex mini", description: "Coding-optimized GPT model for repository edits, reviews, and agentic software work", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-11-13", last_updated: "2025-11-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-5-codex": { id: "openai/gpt-5-codex", name: "GPT-5-Codex", description: "Coding-optimized GPT model for repository edits, reviews, and agentic software work", family: "gpt-codex", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-09-15", last_updated: "2025-09-15", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 38.9, metric: "index", source: "https://openrouter.ai/openai/gpt-5-codex/benchmarks", date: "2026-06-01" }, { name: "SciCode", score: 40.9, metric: "percent correct", source: "https://openrouter.ai/openai/gpt-5-codex/benchmarks", date: "2026-06-01" }, { name: "Terminal-Bench Hard", score: 37.9, metric: "success rate", source: "https://openrouter.ai/openai/gpt-5-codex/benchmarks", date: "2026-06-01" }] }, "openai/gpt-5-mini": { id: "openai/gpt-5-mini", name: "GPT-5 Mini", description: "Small GPT-5 for responsive agents, coding help, and everyday automation", family: "gpt-mini", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-05-30", release_date: "2025-08-07", last_updated: "2025-08-07", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-5.6-luna": { id: "openai/gpt-5.6-luna", name: "GPT-5.6 Luna", description: "Cost-efficient GPT-5.6 model for fast, high-volume workloads", family: "gpt-luna", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2026-02-16", release_date: "2026-07-09", last_updated: "2026-07-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 62.7, metric: "resolve rate", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Terminal-Bench", score: 84.7, metric: "success rate", version: "2.1", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "DeepSWE", score: 67.2, metric: "resolve rate", version: "1.1", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "GPQA Diamond", score: 92.3, metric: "accuracy", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "FrontierMath", score: 78.6, metric: "accuracy", dataset: "Tier 1-3", version: "v2", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "BrowseComp", score: 83.3, metric: "accuracy", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "OSWorld", score: 45.6, metric: "success rate", version: "2.0", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "MMMU Pro", score: 78.4, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Agents' Last Exam", score: 50.3, source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Toolathlon", score: 53.4, metric: "success rate", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Artificial Analysis Intelligence Index", score: 51.2, metric: "index score", variant: "max", version: "4.1", source: "https://artificialanalysis.ai/articles/gpt-5-6-has-landed", date: "2026-07-09" }, { name: "Artificial Analysis Coding Agent Index", score: 74.6, metric: "index score", harness: "Codex", variant: "max", version: "1.1", source: "https://artificialanalysis.ai/articles/gpt-5-6-has-landed", date: "2026-07-09" }] }, "openai/gpt-5.3-codex": { id: "openai/gpt-5.3-codex", name: "GPT-5.3 Codex", description: "Coding-optimized GPT model for repository edits, reviews, and agentic software work", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-08-31", release_date: "2026-02-05", last_updated: "2026-02-05", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "SWE-Atlas Codebase QnA", score: 32.6, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 42.38, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 38.98, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }] }, "openai/gpt-oss-120b": { id: "openai/gpt-oss-120b", name: "GPT OSS 120B", description: "Open GPT reasoning model for self-hosted agents and controllable deployments", family: "gpt-oss", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2025-08-05", last_updated: "2025-08-05", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/openai/gpt-oss-120b" }] }, "openai/gpt-5.3-codex-spark": { id: "openai/gpt-5.3-codex-spark", name: "GPT-5.3 Codex Spark", description: "Coding-optimized GPT model for repository edits, reviews, and agentic software work", family: "gpt-codex-spark", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-08-31", release_date: "2026-02-05", last_updated: "2026-02-05", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 128e3, input: 1e5, output: 32e3 } }, "openai/gpt-4.1-nano": { id: "openai/gpt-4.1-nano", name: "GPT-4.1 nano", description: "Tiny GPT-4.1 option for classification, routing, and very high-volume tasks", family: "gpt-nano", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-04", release_date: "2025-04-14", last_updated: "2025-04-14", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 1047576, output: 32768 }, benchmarks: [{ name: "Aider Polyglot", score: 8.9, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-04-14" }] }, "openai/o1": { id: "openai/o1", name: "o1", description: "O-series reasoning model for hard analysis, math, coding, and planning", family: "o", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2023-09", release_date: "2024-12-05", last_updated: "2024-12-05", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 }, benchmarks: [{ name: "Aider Polyglot", score: 61.7, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2024-12-21" }] }, "openai/gpt-5.2": { id: "openai/gpt-5.2", name: "GPT-5.2", description: "Reliable GPT generation for broad coding, writing, and tool-assisted product work", family: "gpt", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-08-31", release_date: "2025-12-11", last_updated: "2025-12-11", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 29.94, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "openai/gpt-image-1": { id: "openai/gpt-image-1", name: "GPT-Image-1", description: "OpenAI image model for production generation, edits, and brand-safe visual workflows", family: "gpt-image", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2025-04-24", last_updated: "2025-04-24", modalities: { input: ["text", "image"], output: ["image"] }, open_weights: false, limit: { context: 0, output: 0 } }, "openai/gpt-realtime-2.1": { id: "openai/gpt-realtime-2.1", name: "GPT-Realtime-2.1", description: "Realtime speech-to-speech model with configurable reasoning, tool use, and robust voice-agent behavior", family: "gpt", attachment: true, reasoning: true, tool_call: true, structured_output: false, temperature: false, knowledge: "2024-09-30", release_date: "2026-07-06", last_updated: "2026-07-06", modalities: { input: ["text", "audio", "image"], output: ["text", "audio"] }, open_weights: false, limit: { context: 128e3, input: 96e3, output: 32e3 } }, "openai/gpt-5.4-mini": { id: "openai/gpt-5.4-mini", name: "GPT-5.4 mini", description: "Strong small GPT for coding subagents, quick tool use, and high-volume work", family: "gpt-mini", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-08-31", release_date: "2026-03-17", last_updated: "2026-03-17", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 54.4, metric: "resolve rate", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Terminal-Bench", score: 60, metric: "accuracy", variant: "reasoning effort xhigh", version: "2.0", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "MCP Atlas", score: 57.7, metric: "score", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Toolathlon", score: 42.9, metric: "score", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "\u03C4\xB2-Bench Telecom", score: 93.4, metric: "accuracy", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "GPQA Diamond", score: 88, metric: "accuracy", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Humanity's Last Exam", score: 41.5, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Humanity's Last Exam", score: 28.2, metric: "accuracy", variant: "without tools", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OSWorld-Verified", score: 72.1, metric: "success rate", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "MMMU Pro", score: 78, metric: "accuracy", variant: "with Python", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "MMMU Pro", score: 76.6, metric: "accuracy", variant: "without tools", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OmniDocBench", score: 0.1263, metric: "overall edit distance", variant: "reasoning effort none", version: "1.5", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OpenAI MRCR", score: 47.7, metric: "accuracy", variant: "8-needle, 64K-128K", version: "v2", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OpenAI MRCR", score: 33.6, metric: "accuracy", variant: "8-needle, 128K-256K", version: "v2", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Graphwalks", score: 76.3, metric: "accuracy", variant: "BFS, 0-128K", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Graphwalks", score: 71.5, metric: "accuracy", variant: "parents, 0-128K", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }] }, "openai/gpt-4o-2024-05-13": { id: "openai/gpt-4o-2024-05-13", name: "GPT-4o (2024-05-13)", description: "GPT model for general reasoning, writing, coding, and tool-assisted tasks", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2023-09", release_date: "2024-05-13", last_updated: "2024-05-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 4096 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 24.2, metric: "index", source: "https://openrouter.ai/openai/gpt-4o-2024-05-13/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 30.9, metric: "percent correct", source: "https://openrouter.ai/openai/gpt-4o-2024-05-13/benchmarks", date: "2026-03-11" }] }, "openai/o3": { id: "openai/o3", name: "o3", description: "Deliberate o-series reasoner for hard math, coding, and multi-step analysis", family: "o", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-05", release_date: "2025-04-16", last_updated: "2025-04-16", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 }, benchmarks: [{ name: "Aider Polyglot", score: 81.3, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-06-25" }] }, "openai/o4-mini-deep-research": { id: "openai/o4-mini-deep-research", name: "o4-mini-deep-research", description: "Research model for long-horizon investigation, synthesis, and analytical reports", family: "o-mini", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2024-05", release_date: "2024-06-26", last_updated: "2024-06-26", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 } }, "openai/gpt-5-pro": { id: "openai/gpt-5-pro", name: "GPT-5 Pro", description: "Higher-accuracy GPT-5 tier for tough analysis, coding reviews, and planning", family: "gpt-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-10-06", last_updated: "2025-10-06", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 272e3 } }, "openai/gpt-5.5": { id: "openai/gpt-5.5", name: "GPT-5.5", description: "Default frontier GPT for coding, computer use, research, and knowledge work", family: "gpt", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-12-01", release_date: "2026-04-23", last_updated: "2026-04-23", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 58.6, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "Terminal-Bench", score: 78.2, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "SWE-Atlas Codebase QnA", score: 45.43, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 44.79, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 42.59, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "Artificial Analysis Coding Agent Index", score: 65.3, metric: "average pass@1", harness: "Codex", variant: "xhigh", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 80.8, metric: "pass@1", harness: "Codex", variant: "xhigh", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 30.9, metric: "pass@1", harness: "Codex", variant: "xhigh", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 84.1, metric: "pass@1", harness: "Codex", variant: "xhigh", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Artificial Analysis Coding Agent Index", score: 60.4, metric: "average pass@1", harness: "Codex", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 79.1, metric: "pass@1", harness: "Codex", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 26.2, metric: "pass@1", harness: "Codex", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 75.8, metric: "pass@1", harness: "Codex", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Artificial Analysis Coding Agent Index", score: 57.8, metric: "average pass@1", harness: "Cursor CLI", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 75, metric: "pass@1", harness: "Cursor CLI", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 24.9, metric: "pass@1", harness: "Cursor CLI", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 73.4, metric: "pass@1", harness: "Cursor CLI", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 82.7, metric: "success rate", version: "2.0", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GPQA Diamond", score: 93.6, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 41.4, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 52.2, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "OSWorld-Verified", score: 78.7, metric: "success rate", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "BrowseComp", score: 84.4, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "MMMU Pro", score: 81.2, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "ARC-AGI-2", score: 85, metric: "accuracy", variant: "Verified", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 51.7, metric: "accuracy", dataset: "Tier 1-3", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 35.4, metric: "accuracy", dataset: "Tier 4", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GDPval", score: 84.9, metric: "wins or ties", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "MCP Atlas", score: 75.3, metric: "success rate", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Toolathlon", score: 55.6, metric: "success rate", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "\u03C4\xB2-Bench Telecom", score: 98, metric: "success rate", variant: "original prompts", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }] }, "openai/gpt-5.5-instant": { id: "openai/gpt-5.5-instant", name: "GPT-5.5 Instant", description: "Compact GPT model for low-latency assistance and high-volume workloads", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-12-01", release_date: "2026-05-05", last_updated: "2026-05-28", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 4e5, output: 128e3 } }, "openai/gpt-5.2-codex": { id: "openai/gpt-5.2-codex", name: "GPT-5.2 Codex", description: "Code-specialist GPT for repository edits, reviews, and long-running software agents", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-08-31", release_date: "2025-12-11", last_updated: "2025-12-11", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 41.04, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "openai/gpt-4.1": { id: "openai/gpt-4.1", name: "GPT-4.1", description: "Long-lived GPT workhorse for coding, instruction following, and production apps", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-04", release_date: "2025-04-14", last_updated: "2025-04-14", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1047576, output: 32768 }, benchmarks: [{ name: "Aider Polyglot", score: 52.4, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-04-14" }] }, "openai/gpt-4o-2024-08-06": { id: "openai/gpt-4o-2024-08-06", name: "GPT-4o (2024-08-06)", description: "GPT model for general reasoning, writing, coding, and tool-assisted tasks", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2023-09", release_date: "2024-08-06", last_updated: "2024-08-06", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 }, benchmarks: [{ name: "Aider Polyglot", score: 23.1, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2024-12-30" }, { name: "Artificial Analysis Coding Index", score: 16.6, metric: "index", source: "https://openrouter.ai/openai/gpt-4o-2024-08-06/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 33.1, metric: "percent correct", source: "https://openrouter.ai/openai/gpt-4o-2024-08-06/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 8.3, metric: "success rate", source: "https://openrouter.ai/openai/gpt-4o-2024-08-06/benchmarks", date: "2026-03-11" }] }, "openai/gpt-5.6-terra": { id: "openai/gpt-5.6-terra", name: "GPT-5.6 Terra", description: "Balanced GPT-5.6 model for capable, cost-efficient everyday work", family: "gpt-terra", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2026-02-16", release_date: "2026-07-09", last_updated: "2026-07-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 63.4, metric: "resolve rate", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Terminal-Bench", score: 87.4, metric: "success rate", version: "2.1", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "DeepSWE", score: 69.6, metric: "resolve rate", version: "1.1", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "GPQA Diamond", score: 92.9, metric: "accuracy", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "FrontierMath", score: 84.9, metric: "accuracy", dataset: "Tier 1-3", version: "v2", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "BrowseComp", score: 87.5, metric: "accuracy", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "OSWorld", score: 50.2, metric: "success rate", version: "2.0", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "MMMU Pro", score: 80.7, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Agents' Last Exam", score: 50.4, source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Toolathlon", score: 53.1, metric: "success rate", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Artificial Analysis Intelligence Index", score: 55, metric: "index score", variant: "max", version: "4.1", source: "https://artificialanalysis.ai/articles/gpt-5-6-has-landed", date: "2026-07-09" }, { name: "Artificial Analysis Coding Agent Index", score: 77.4, metric: "index score", harness: "Codex", variant: "max", version: "1.1", source: "https://artificialanalysis.ai/articles/gpt-5-6-has-landed", date: "2026-07-09" }] }, "openai/o4-mini": { id: "openai/o4-mini", name: "o4-mini", description: "Fast o-series model for compact reasoning, coding, and tool use", family: "o-mini", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-05", release_date: "2025-04-16", last_updated: "2025-04-16", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 }, benchmarks: [{ name: "Aider Polyglot", score: 72, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-04-16" }] }, "openai/gpt-5.4": { id: "openai/gpt-5.4", name: "GPT-5.4", description: "Agent-ready GPT for coding and computer-use workflows at a lower cost", family: "gpt", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-08-31", release_date: "2026-03-05", last_updated: "2026-03-05", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 59.1, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }, { name: "SWE-Atlas Codebase QnA", score: 40.8, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Codebase QnA", score: 36.3, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 44.29, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 44.36, metric: "score", harness: "Codex CLI", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "SWE-Atlas Test Writing", score: 40, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "Artificial Analysis Coding Agent Index", score: 53.6, metric: "average pass@1", harness: "Codex", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 72.4, metric: "pass@1", harness: "Codex", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 18.4, metric: "pass@1", harness: "Codex", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 69.8, metric: "pass@1", harness: "Codex", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Artificial Analysis Coding Agent Index", score: 52.2, metric: "average pass@1", harness: "Cursor CLI", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 72.9, metric: "pass@1", harness: "Cursor CLI", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 18.9, metric: "pass@1", harness: "Cursor CLI", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 64.7, metric: "pass@1", harness: "Cursor CLI", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 75.1, metric: "success rate", version: "2.0", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GPQA Diamond", score: 92.8, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 39.8, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 52.1, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "OSWorld-Verified", score: 75, metric: "success rate", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "BrowseComp", score: 82.7, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GDPval", score: 83, metric: "wins or ties", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "ARC-AGI-2", score: 73.3, metric: "accuracy", variant: "Verified", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 47.6, metric: "accuracy", dataset: "Tier 1-3", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 27.1, metric: "accuracy", dataset: "Tier 4", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "MMMU Pro", score: 81.2, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }] }, "openai/o3-mini": { id: "openai/o3-mini", name: "o3-mini", description: "Smaller o-series reasoner for economical coding, math, and planning tasks", family: "o-mini", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-05", release_date: "2024-12-20", last_updated: "2025-01-29", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 }, benchmarks: [{ name: "Aider Polyglot", score: 60.4, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-01-31" }] }, "openai/gpt-4o-2024-11-20": { id: "openai/gpt-4o-2024-11-20", name: "GPT-4o (2024-11-20)", description: "GPT model for general reasoning, writing, coding, and tool-assisted tasks", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2023-09", release_date: "2024-11-20", last_updated: "2024-11-20", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 }, benchmarks: [{ name: "Aider Polyglot", score: 18.2, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2024-12-30" }, { name: "Artificial Analysis Coding Index", score: 16.7, metric: "index", source: "https://openrouter.ai/openai/gpt-4o-2024-11-20/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 33.3, metric: "percent correct", source: "https://openrouter.ai/openai/gpt-4o-2024-11-20/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 8.3, metric: "success rate", source: "https://openrouter.ai/openai/gpt-4o-2024-11-20/benchmarks", date: "2026-03-11" }] }, "openai/gpt-oss-safeguard-120b": { id: "openai/gpt-oss-safeguard-120b", name: "GPT OSS Safeguard 120B", description: "Safety model for policy screening, moderation, and risk-aware routing workflows", family: "gpt-oss", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2025-10-29", last_updated: "2025-10-29", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/openai/gpt-oss-safeguard-120b" }] }, "openai/gpt-realtime-whisper": { id: "openai/gpt-realtime-whisper", name: "GPT Realtime Whisper", description: "Streaming speech-to-text model for low-latency transcript deltas from live audio", family: "whisper", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2026-05-07", last_updated: "2026-05-07", modalities: { input: ["audio"], output: ["text"] }, open_weights: false, limit: { context: 0, output: 0 } }, "openai/gpt-5.2-chat-latest": { id: "openai/gpt-5.2-chat-latest", name: "GPT-5.2 Chat", description: "Chat-tuned GPT model for conversational assistance, writing, and tool workflows", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-08-31", release_date: "2025-12-11", last_updated: "2025-12-11", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 } }, "openai/gpt-5.2-pro": { id: "openai/gpt-5.2-pro", name: "GPT-5.2 Pro", description: "Higher-accuracy GPT-5.2 variant for tougher reasoning and review workflows", family: "gpt-pro", attachment: true, reasoning: true, tool_call: true, structured_output: false, temperature: false, knowledge: "2025-08-31", release_date: "2025-12-11", last_updated: "2025-12-11", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-5.1-chat-latest": { id: "openai/gpt-5.1-chat-latest", name: "GPT-5.1 Chat", description: "Chat-tuned GPT-5.1 for polished assistants, writing, and product conversations", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-11-13", last_updated: "2025-11-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 } }, "openai/gpt-image-2": { id: "openai/gpt-image-2", name: "GPT-Image-2", description: "Image model for prompt-driven generation, editing, and visual design workflows", family: "gpt-image", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2026-04-21", last_updated: "2026-04-21", modalities: { input: ["text", "image"], output: ["image"] }, open_weights: false, limit: { context: 0, output: 0 } }, "openai/gpt-5.1-codex-max": { id: "openai/gpt-5.1-codex-max", name: "GPT-5.1 Codex Max", description: "Coding-optimized GPT model for repository edits, reviews, and agentic software work", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-11-13", last_updated: "2025-11-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-5.1-codex": { id: "openai/gpt-5.1-codex", name: "GPT-5.1 Codex", description: "Codex GPT for repository edits, code review, and practical software agents", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-11-13", last_updated: "2025-11-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-4.1-mini": { id: "openai/gpt-4.1-mini", name: "GPT-4.1 mini", description: "Affordable GPT-4.1 lane for fast coding help and structured extraction", family: "gpt-mini", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-04", release_date: "2025-04-14", last_updated: "2025-04-14", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1047576, output: 32768 }, benchmarks: [{ name: "Aider Polyglot", score: 32.4, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-04-14" }] }, "openai/gpt-oss-20b": { id: "openai/gpt-oss-20b", name: "GPT OSS 20B", description: "Open GPT reasoning model for self-hosted agents and controllable deployments", family: "gpt-oss", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2025-08-05", last_updated: "2025-08-05", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/openai/gpt-oss-20b" }] }, "openai/whisper-large-v3-turbo": { id: "openai/whisper-large-v3-turbo", name: "Whisper Large v3 Turbo", description: "Speech transcription model for accurate audio-to-text and captioning workflows", family: "whisper", attachment: false, reasoning: false, tool_call: false, release_date: "2024-10-01", last_updated: "2024-10-01", modalities: { input: ["audio"], output: ["text"] }, open_weights: true, limit: { context: 448, output: 448 } }, "openai/gpt-5.4-nano": { id: "openai/gpt-5.4-nano", name: "GPT-5.4 nano", description: "Cheapest GPT-5.4 lane for simple routing, extraction, and bulk automation", family: "gpt-nano", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-08-31", release_date: "2026-03-17", last_updated: "2026-03-17", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 52.4, metric: "resolve rate", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Terminal-Bench", score: 46.3, metric: "accuracy", variant: "reasoning effort xhigh", version: "2.0", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "MCP Atlas", score: 56.1, metric: "score", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Toolathlon", score: 35.5, metric: "score", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "\u03C4\xB2-Bench Telecom", score: 92.5, metric: "accuracy", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "GPQA Diamond", score: 82.8, metric: "accuracy", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Humanity's Last Exam", score: 37.7, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Humanity's Last Exam", score: 24.3, metric: "accuracy", variant: "without tools", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OSWorld-Verified", score: 39, metric: "success rate", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "MMMU Pro", score: 69.5, metric: "accuracy", variant: "with Python", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "MMMU Pro", score: 66.1, metric: "accuracy", variant: "without tools", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OmniDocBench", score: 0.2419, metric: "overall edit distance", variant: "reasoning effort none", version: "1.5", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OpenAI MRCR", score: 44.2, metric: "accuracy", variant: "8-needle, 64K-128K", version: "v2", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OpenAI MRCR", score: 33.1, metric: "accuracy", variant: "8-needle, 128K-256K", version: "v2", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Graphwalks", score: 73.4, metric: "accuracy", variant: "BFS, 0-128K", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Graphwalks", score: 50.8, metric: "accuracy", variant: "parents, 0-128K", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }] }, "openai/gpt-5.1": { id: "openai/gpt-5.1", name: "GPT-5.1", description: "Sharper GPT-5 generation for coding, product work, and tool-assisted tasks", family: "gpt", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-11-13", last_updated: "2025-11-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "xai/grok-4.6": { id: "xai/grok-4.6", name: "Grok 4.6", description: "xAI's frontier model for long-running agents, coding, knowledge work, and visual projects", family: "grok", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-02-01", release_date: "2026-08-12", last_updated: "2026-08-12", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 5e5, output: 5e5 } }, "xai/grok-4.5": { id: "xai/grok-4.5", name: "Grok 4.5", description: "xAI's Grok model for chat, coding, agentic tools, and lower hallucination risk", family: "grok", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-07-08", last_updated: "2026-07-08", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 5e5, output: 5e5 }, benchmarks: [{ name: "SWE-Bench Pro", score: 64.7, metric: "resolve rate", source: "https://x.ai/news/grok-4-5", date: "2026-07-08" }, { name: "SWE-Bench Multilingual", score: 78, metric: "resolve rate", source: "https://x.ai/news/grok-4-5", date: "2026-07-08" }, { name: "Terminal-Bench", score: 83.3, metric: "success rate", version: "2.1", source: "https://x.ai/news/grok-4-5", date: "2026-07-08" }, { name: "DeepSWE", score: 62, metric: "resolve rate", version: "1.0", source: "https://x.ai/news/grok-4-5", date: "2026-07-08" }, { name: "DeepSWE", score: 53, metric: "resolve rate", harness: "mini-swe-agent", version: "1.1", source: "https://x.ai/news/grok-4-5", date: "2026-07-08" }, { name: "SWE Marathon", score: 29, metric: "pass@1", source: "https://x.ai/news/grok-4-5", date: "2026-07-08" }] }, "xai/grok-4.20-0309-non-reasoning": { id: "xai/grok-4.20-0309-non-reasoning", name: "Grok 4.20 (Non-Reasoning)", description: "Grok model for agentic tool use, reasoning, coding, and live assistance", family: "grok", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2026-03-09", last_updated: "2026-03-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 3e4 } }, "xai/grok-imagine-video-1.5": { id: "xai/grok-imagine-video-1.5", name: "Grok Imagine Video 1.5", description: "Video model for image-to-video generation, editing, and extension workflows", family: "grok", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2026-05-30", last_updated: "2026-05-30", modalities: { input: ["text", "image", "video"], output: ["video"] }, open_weights: false, limit: { context: 1024, output: 0 } }, "xai/grok-4.1-fast": { id: "xai/grok-4.1-fast", name: "Grok 4.1 Fast", description: "xAI's fast agentic tool-calling model with a 2M context window; non-reasoning variant for low-latency responses", family: "grok", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2025-11-19", last_updated: "2025-11-19", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e6, output: 3e4 } }, "xai/grok-4.20-0309-reasoning": { id: "xai/grok-4.20-0309-reasoning", name: "Grok 4.20 (Reasoning)", description: "Reasoning Grok for document-heavy analysis and long-horizon tool use", family: "grok", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-03-09", last_updated: "2026-03-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 3e4 } }, "xai/grok-imagine-image-2.0": { id: "xai/grok-imagine-image-2.0", name: "Grok Imagine Image 2.0", description: "Image model for prompt-driven generation, editing, and visual design workflows", family: "grok", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2026-08-07", last_updated: "2026-08-07", modalities: { input: ["text", "image"], output: ["image"] }, open_weights: false, limit: { context: 8e3, output: 0 } }, "xai/grok-build-0.1": { id: "xai/grok-build-0.1", name: "Grok Build 0.1", description: "Fast Grok coding model tuned for agentic engineering and iterative edits", family: "grok-build", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-16", last_updated: "2026-04-16", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 256e3 } }, "xai/grok-4.3": { id: "xai/grok-4.3", name: "Grok 4.3", description: "xAI's default Grok for chat, coding, agentic tools, and lower hallucination risk", family: "grok", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-17", last_updated: "2026-04-17", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 3e4 }, benchmarks: [{ name: "Artificial Analysis Intelligence Index", score: 53, metric: "index score", version: "4.0", source: "https://artificialanalysis.ai/articles/xai-launches-grok-4-3-with-improved-agentic-performance-and-lower-pricing", date: "2026-04-30" }, { name: "GDPval-AA", score: 1500, metric: "Elo", source: "https://artificialanalysis.ai/articles/xai-launches-grok-4-3-with-improved-agentic-performance-and-lower-pricing", date: "2026-04-30" }, { name: "\u03C4\xB2-Bench Telecom", score: 98, metric: "success rate", source: "https://artificialanalysis.ai/articles/xai-launches-grok-4-3-with-improved-agentic-performance-and-lower-pricing", date: "2026-04-30" }, { name: "IFBench", score: 81, metric: "accuracy", source: "https://artificialanalysis.ai/articles/xai-launches-grok-4-3-with-improved-agentic-performance-and-lower-pricing", date: "2026-04-30" }] }, "meituan/longcat-2.0": { id: "meituan/longcat-2.0", name: "LongCat-2.0", description: "Meituan LongCat-2.0, a reasoning model with tool calling and a 1M-token context window", family: "longcat", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-30", last_updated: "2026-06-30", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 131072 }, benchmarks: [{ name: "SWE-Bench Pro", score: 59.5, metric: "resolve rate", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }, { name: "SWE-Bench Multilingual", score: 77.3, metric: "resolve rate", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }, { name: "Terminal-Bench", score: 70.8, metric: "success rate", version: "2.1", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }, { name: "GPQA Diamond", score: 88.9, metric: "accuracy", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }, { name: "BrowseComp", score: 79.9, metric: "accuracy", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }, { name: "IFEval", score: 90, metric: "accuracy", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }, { name: "FORTE", score: 73.2, metric: "success rate", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }] }, "swiss-ai/apertus-70b": { id: "swiss-ai/apertus-70b", name: "Apertus 70B", description: "Fully open 70B multilingual LLM supporting 1800+ languages with 65K context. Trained on 15T tokens of compliant open data. Apache 2.0, EU AI Act compliant.", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-09", release_date: "2025-09-02", last_updated: "2025-09-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 65536, output: 8192 }, license: "Apache-2.0", links: [{ label: "Paper", url: "https://arxiv.org/abs/2509.14233", type: "paper" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509" }] }, "swiss-ai/apertus-8b": { id: "swiss-ai/apertus-8b", name: "Apertus 8B", description: "Fully open 8B multilingual LLM supporting 1800+ languages with 65K context. Trained on compliant open data. Apache 2.0, EU AI Act compliant.", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-09", release_date: "2025-09-02", last_updated: "2025-09-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 65536, output: 8192 }, license: "Apache-2.0", links: [{ label: "Paper", url: "https://arxiv.org/abs/2509.14233", type: "paper" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/swiss-ai/Apertus-8B-Instruct-2509" }] }, "stepfun/step-3.5-flash-2603": { id: "stepfun/step-3.5-flash-2603", name: "Step 3.5 Flash 2603", description: "StepFun flash model for efficient multimodal reasoning, coding, and tool use", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-01", release_date: "2026-04-02", last_updated: "2026-04-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, input: 256e3, output: 256e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/stepfun-ai/Step-3.5-Flash" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 34.6, metric: "index", source: "https://openrouter.ai/stepfun/step-3.5-flash/benchmarks", date: "2026-06-02" }, { name: "SciCode", score: 38.5, metric: "percent correct", source: "https://openrouter.ai/stepfun/step-3.5-flash/benchmarks", date: "2026-06-02" }, { name: "Terminal-Bench Hard", score: 32.6, metric: "success rate", source: "https://openrouter.ai/stepfun/step-3.5-flash/benchmarks", date: "2026-06-02" }] }, "stepfun/step-3.5-flash": { id: "stepfun/step-3.5-flash", name: "Step 3.5 Flash", description: "StepFun flash lane for quick multimodal reasoning and coding assistance", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-01", release_date: "2026-01-29", last_updated: "2026-02-13", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, input: 256e3, output: 256e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/stepfun-ai/Step-3.5-Flash" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 31.6, metric: "index", source: "https://openrouter.ai/stepfun/step-3.5-flash/benchmarks", date: "2026-06-02" }, { name: "SciCode", score: 40.4, metric: "percent correct", source: "https://openrouter.ai/stepfun/step-3.5-flash/benchmarks", date: "2026-06-02" }, { name: "Terminal-Bench Hard", score: 27.3, metric: "success rate", source: "https://openrouter.ai/stepfun/step-3.5-flash/benchmarks", date: "2026-06-02" }, { name: "SWE-Bench Verified", score: 74.4, metric: "resolved", source: "https://arxiv.org/abs/2602.10604" }] }, "stepfun/step-3.7-flash": { id: "stepfun/step-3.7-flash", name: "Step 3.7 Flash", description: "Newer StepFun flash model for faster agents, coding, and multimodal prompts", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2026-03-01", release_date: "2026-05-29", last_updated: "2026-05-29", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 256e3, input: 256e3, output: 256e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/stepfun-ai/Step-3.7-Flash" }], benchmarks: [{ name: "SWE-Bench Pro", score: 56.3, metric: "resolve rate", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "SWE-Bench Verified", score: 76.5, metric: "resolved", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "Terminal-Bench", score: 59.6, metric: "success rate", version: "2.1", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "Humanity's Last Exam", score: 47.2, metric: "accuracy", variant: "with tools", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "BrowseComp", score: 75.8, metric: "accuracy", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "Toolathlon", score: 49.5, metric: "success rate", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "GDPval", score: 45.8, metric: "wins or ties", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "ClawEval", score: 67.1, metric: "pass^3", version: "1.1", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "Artificial Analysis Coding Index", score: 37.1, metric: "index", source: "https://openrouter.ai/stepfun/step-3.7-flash/benchmarks", date: "2026-06-15" }, { name: "SciCode", score: 40, metric: "percent correct", source: "https://openrouter.ai/stepfun/step-3.7-flash/benchmarks", date: "2026-06-15" }, { name: "Terminal-Bench Hard", score: 35.6, metric: "success rate", source: "https://openrouter.ai/stepfun/step-3.7-flash/benchmarks", date: "2026-06-15" }] }, "sarvam/sarvam-105b": { id: "sarvam/sarvam-105b", name: "Sarvam 105B", description: "Flagship Indian-language reasoning model for enterprise multilingual applications", family: "sarvam", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-09-01", last_updated: "2025-09-01", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 } }, "sarvam/sarvam-30b": { id: "sarvam/sarvam-30b", name: "Sarvam 30B", description: "Efficient Indian-language reasoning model for chat, coding, and multilingual work", family: "sarvam", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-02-18", last_updated: "2026-02-18", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 } }, "trendyol/asure-12b": { id: "trendyol/asure-12b", name: "Trendyol Asure 12B", description: "Turkish-language multimodal instruct model built on Gemma 3 12B for e-commerce text, chat, and image-text tasks", family: "gemma", attachment: true, reasoning: false, tool_call: false, temperature: true, release_date: "2026-02-19", last_updated: "2026-02-20", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072 }, license: "Gemma", weights: [{ label: "Hugging Face", url: "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B" }] }, "deepreinforce/ornith-1.0-35b": { id: "deepreinforce/ornith-1.0-35b", name: "Ornith 1.0 35B", description: "Large coding-reasoning model for agentic software tasks and RL search", family: "ornith", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-25", last_updated: "2026-06-25", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144 }, license: "MIT", links: [{ label: "Model card", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B", type: "model_card" }, { label: "Announcement", url: "https://deep-reinforce.com/ornith_1_0.html", type: "announcement" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 75.6, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }, { name: "SWE-Bench Pro", score: 50.4, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }, { name: "SWE-Bench Multilingual", score: 69.3, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }, { name: "Terminal-Bench 2.1", score: 64.2, metric: "percent", variant: "Terminus-2", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }, { name: "Terminal-Bench 2.1", score: 62.8, metric: "percent", variant: "Claude Code", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }, { name: "NL2Repo", score: 34.6, metric: "percent", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }, { name: "Claw-eval", score: 69.8, metric: "percent", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }] }, "deepreinforce/ornith-1.0-397b": { id: "deepreinforce/ornith-1.0-397b", name: "Ornith 1.0 397B", description: "Large coding-reasoning model for agentic software tasks and RL search", family: "ornith", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-25", last_updated: "2026-06-25", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144 }, license: "MIT", links: [{ label: "Model card", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B", type: "model_card" }, { label: "Announcement", url: "https://deep-reinforce.com/ornith_1_0.html", type: "announcement" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { label: "Hugging Face (FP8)", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B-FP8", quantization: "fp8" }], benchmarks: [{ name: "SWE-Bench Verified", score: 82.4, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { name: "SWE-Bench Pro", score: 62.2, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { name: "SWE-Bench Multilingual", score: 78.9, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { name: "Terminal-Bench 2.1", score: 77.5, metric: "percent", variant: "Terminus-2", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { name: "Terminal-Bench 2.1", score: 78.2, metric: "percent", variant: "Claude Code", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { name: "NL2Repo", score: 48.2, metric: "percent", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { name: "Claw-eval", score: 77.1, metric: "percent", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }] }, "deepreinforce/ornith-1.0-31b": { id: "deepreinforce/ornith-1.0-31b", name: "Ornith 1.0 31B", description: "Open coding-reasoning model for repository tasks and self-improving agents", family: "ornith", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-25", last_updated: "2026-06-25", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144 }, license: "MIT", links: [{ label: "Announcement", url: "https://deep-reinforce.com/ornith_1_0.html", type: "announcement" }] }, "deepreinforce/ornith-1.0-9b": { id: "deepreinforce/ornith-1.0-9b", name: "Ornith 1.0 9B", description: "Open coding-reasoning model for repository tasks and self-improving agents", family: "ornith", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-25", last_updated: "2026-06-25", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144 }, license: "MIT", links: [{ label: "Model card", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B", type: "model_card" }, { label: "Announcement", url: "https://deep-reinforce.com/ornith_1_0.html", type: "announcement" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 69.4, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }, { name: "SWE-Bench Pro", score: 42.9, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }, { name: "SWE-Bench Multilingual", score: 52, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }, { name: "Terminal-Bench 2.1", score: 43.1, metric: "percent", variant: "Terminus-2", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }, { name: "Terminal-Bench 2.1", score: 40.6, metric: "percent", variant: "Claude Code", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }, { name: "NL2Repo", score: 27.2, metric: "percent", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }, { name: "Claw-eval", score: 63.1, metric: "percent", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }] }, "xiaomi/mimo-v2.5-pro-ultraspeed": { id: "xiaomi/mimo-v2.5-pro-ultraspeed", name: "MiMo-V2.5-Pro-UltraSpeed", description: "MiMo pro model for strong multimodal reasoning and agent execution", family: "mimo", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-12", release_date: "2026-06-08", last_updated: "2026-06-09", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro-FP4-DFlash" }] }, "xiaomi/mimo-v2.5": { id: "xiaomi/mimo-v2.5", name: "MiMo-V2.5", description: "Open MiMo model for multimodal coding agents and long-context automation", family: "mimo", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-12", release_date: "2026-04-22", last_updated: "2026-04-22", modalities: { input: ["text", "image", "audio", "video"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/XiaomiMiMo/MiMo-V2.5" }] }, "xiaomi/mimo-v2-pro": { id: "xiaomi/mimo-v2-pro", name: "MiMo-V2-Pro", description: "Earlier MiMo Pro model for multimodal agents, reasoning, and code tasks", family: "mimo", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-12", release_date: "2026-03-18", last_updated: "2026-03-18", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 131072 } }, "xiaomi/mimo-v2-omni": { id: "xiaomi/mimo-v2-omni", name: "MiMo-V2-Omni", description: "MiMo omni model for text, image, video, audio, and agents", family: "mimo", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-12", release_date: "2026-03-18", last_updated: "2026-03-18", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 131072 } }, "xiaomi/mimo-v2-flash": { id: "xiaomi/mimo-v2-flash", name: "MiMo-V2-Flash", description: "MiMo flash model for fast multimodal assistance and agent workflows", family: "mimo", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-12-01", release_date: "2025-12-16", last_updated: "2026-02-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash" }] }, "xiaomi/mimo-v2.5-pro": { id: "xiaomi/mimo-v2.5-pro", name: "MiMo-V2.5-Pro", description: "Stronger MiMo Pro tier for multimodal reasoning and coding-agent execution", family: "mimo", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-12", release_date: "2026-04-22", last_updated: "2026-04-22", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro" }], benchmarks: [{ name: "SWE-Bench Verified", score: 78.9, metric: "resolved", source: "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro" }, { name: "SWE-Bench Pro", score: 57.2, metric: "resolve rate", source: "https://mimo.xiaomi.com/mimo-v2-5-pro/", date: "2026-04-22" }, { name: "GPQA Diamond", score: 86.6, metric: "accuracy", source: "https://mimo.xiaomi.com/mimo-v2-5-pro/", date: "2026-04-22" }] }, "alibaba/qwen3-vl-plus": { id: "alibaba/qwen3-vl-plus", name: "Qwen3-VL Plus", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-09-23", last_updated: "2025-09-23", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 32768 } }, "alibaba/qwen3.8-max-preview": { id: "alibaba/qwen3.8-max-preview", name: "Qwen3.8 Max Preview", description: "Preview Qwen flagship for million-token multimodal reasoning and long-horizon agentic workflows", family: "qwen", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-07-19", last_updated: "2026-07-19", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 131072 }, benchmarks: [{ name: "Terminal-Bench", score: 86.6, metric: "accuracy", variant: "xhigh", version: "2.1", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "SWE-Bench Pro", score: 67.7, metric: "resolve rate", harness: "Claude Code", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "DeepSWE", score: 56.6, metric: "resolve rate", harness: "Claude Code", variant: "xhigh", version: "1.1", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "NL2Repo", score: 55.9, metric: "resolve rate", harness: "Claude Code", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "FrontierSWE", score: 73.5, metric: "dominance score", harness: "Claude Code", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "MLS-Bench-Lite", score: 41, metric: "score", harness: "Claude Code", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "AutomationBench", score: 27.3, metric: "pass@1", variant: "xhigh", dataset: "600-task public subset", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "Toolathlon Verified", score: 72.5, metric: "pass@1", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "WideSearch", score: 81.9, metric: "F1", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "Humanity's Last Exam", score: 56.2, metric: "accuracy", variant: "xhigh, with tools", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "GPQA Diamond", score: 92.6, metric: "accuracy", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "Humanity's Last Exam", score: 43.6, metric: "accuracy", variant: "xhigh, no tools", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "IFBench", score: 82.8, metric: "score", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "OSWorld-Verified", score: 86.1, metric: "success rate", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "MMMU Pro", score: 82.3, metric: "accuracy", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }] }, "alibaba/qwen3-coder-30b-a3b-instruct": { id: "alibaba/qwen3-coder-30b-a3b-instruct", name: "Qwen3-Coder 30B-A3B Instruct", description: "Smaller Qwen coder for efficient local agents and repo-level fixes", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-04", last_updated: "2025-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 19.4, metric: "index", source: "https://openrouter.ai/qwen/qwen3-coder-30b-a3b-instruct/benchmarks", date: "2026-06-02" }, { name: "SciCode", score: 27.8, metric: "percent correct", source: "https://openrouter.ai/qwen/qwen3-coder-30b-a3b-instruct/benchmarks", date: "2026-06-02" }, { name: "Terminal-Bench Hard", score: 15.2, metric: "success rate", source: "https://openrouter.ai/qwen/qwen3-coder-30b-a3b-instruct/benchmarks", date: "2026-06-02" }] }, "alibaba/qwen3.7-max": { id: "alibaba/qwen3.7-max", name: "Qwen3.7 Max", description: "Qwen frontier model tuned for agent frameworks, coding assistants, and long tasks", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-05-21", last_updated: "2026-05-21", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 }, benchmarks: [{ name: "SWE-Bench Verified", score: 80.4, metric: "resolved", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "SWE-Bench Pro", score: 60.6, metric: "resolve rate", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "SWE-Bench Multilingual", score: 78.3, metric: "resolve rate", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "Terminal-Bench", score: 69.7, metric: "success rate", harness: "Terminus-2", version: "2.0", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "GPQA Diamond", score: 92.4, metric: "accuracy", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "Humanity's Last Exam", score: 41.4, metric: "accuracy", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "SciCode", score: 53.5, source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "MCP Atlas", score: 76.4, metric: "success rate", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "NL2Repo", score: 47.2, harness: "Claude Code", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }] }, "alibaba/qwen-turbo": { id: "alibaba/qwen-turbo", name: "Qwen Turbo", description: "Efficient Qwen model for fast chat, extraction, and high-volume workloads", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2024-11-01", last_updated: "2025-04-28", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 16384 } }, "alibaba/qwen-omni-turbo": { id: "alibaba/qwen-omni-turbo", name: "Qwen-Omni Turbo", description: "Qwen omni model for text, vision, audio, and multimodal agent tasks", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2025-01-19", last_updated: "2025-03-26", modalities: { input: ["text", "image", "audio", "video"], output: ["text", "audio"] }, open_weights: false, limit: { context: 32768, output: 2048 } }, "alibaba/qwen-vl-max": { id: "alibaba/qwen-vl-max", name: "Qwen-VL Max", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2024-04-08", last_updated: "2025-08-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 8192 } }, "alibaba/qwen3-32b": { id: "alibaba/qwen3-32b", name: "Qwen3 32B", description: "Dense open Qwen model for self-hosted chat, reasoning, and coding", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-04", last_updated: "2025-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-32B" }], benchmarks: [{ name: "Aider Polyglot", score: 40, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-08" }] }, "alibaba/qwen3-235b-a22b": { id: "alibaba/qwen3-235b-a22b", name: "Qwen3 235B-A22B", description: "Large open Qwen MoE for multilingual reasoning, coding, and tool use", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-04", last_updated: "2025-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-235B-A22B" }], benchmarks: [{ name: "Aider Polyglot", score: 59.6, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-09" }, { name: "SWE-Bench Pro", score: 21.41, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "alibaba/qwen-max": { id: "alibaba/qwen-max", name: "Qwen Max", description: "Flagship Qwen model for complex reasoning, coding, and agentic workflows", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2024-04-03", last_updated: "2025-01-25", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 32768, output: 8192 }, benchmarks: [{ name: "Aider Polyglot", score: 21.8, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-01-28" }] }, "alibaba/qwen3.5-35b-a3b": { id: "alibaba/qwen3.5-35b-a3b", name: "Qwen3.5 35B-A3B", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-23", last_updated: "2026-02-23", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.5-35B-A3B" }] }, "alibaba/qwen3-vl-235b-a22b-thinking": { id: "alibaba/qwen3-vl-235b-a22b-thinking", name: "Qwen3 VL 235B A22B Thinking", description: "Qwen vision-language thinking model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-09-23", last_updated: "2025-09-23", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Thinking" }] }, "alibaba/qwen3-30b-a3b": { id: "alibaba/qwen3-30b-a3b", name: "Qwen3 30B A3B", description: "Sparse MoE Qwen model with 3B active parameters for efficient chat and reasoning", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-04-28", last_updated: "2025-04-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-30B-A3B" }] }, "alibaba/qwen3.5-397b-a17b": { id: "alibaba/qwen3.5-397b-a17b", name: "Qwen3.5 397B-A17B", description: "Large open Qwen multimodal MoE for visual agents and long technical tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-15", last_updated: "2026-02-15", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.5-397B-A17B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 76.4, metric: "resolved", source: "https://huggingface.co/Qwen/Qwen3.5-397B-A17B" }] }, "alibaba/qwen3.5-flash": { id: "alibaba/qwen3.5-flash", name: "Qwen3.5 Flash", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-23", last_updated: "2026-02-23", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "alibaba/qwen2.5-coder-0.5b": { id: "alibaba/qwen2.5-coder-0.5b", name: "Qwen2.5-Coder-0.5B", description: "Tiny open Qwen code model for lightweight completion and on-device coding", family: "qwen", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2024-11-12", last_updated: "2024-11-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 32768, output: 8192 }, license: "Apache 2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B" }] }, "alibaba/qwen-plus": { id: "alibaba/qwen-plus", name: "Qwen Plus", description: "Qwen instruction model for multilingual chat, reasoning, and tool use", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2024-01-25", last_updated: "2025-09-11", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 32768 } }, "alibaba/qwen3-coder-next": { id: "alibaba/qwen3-coder-next", name: "Qwen3 Coder Next", description: "Open-weight Qwen coding model for agents, repository edits, and multi-turn tool use", family: "qwen", attachment: false, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-09", release_date: "2026-02-03", last_updated: "2026-02-03", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-Coder-Next" }] }, "alibaba/qwen3.5-plus": { id: "alibaba/qwen3.5-plus", name: "Qwen3.5 Plus", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2026-02-16", last_updated: "2026-02-16", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "alibaba/qwen2-5-vl-72b-instruct": { id: "alibaba/qwen2-5-vl-72b-instruct", name: "Qwen2.5-VL 72B Instruct", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2024-09", last_updated: "2024-09", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct" }] }, "alibaba/qwen3-coder-flash": { id: "alibaba/qwen3-coder-flash", name: "Qwen3 Coder Flash", description: "Qwen coding model for software agents, repository edits, and code reasoning", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-07-28", last_updated: "2025-07-28", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "alibaba/qwen3.6-max-preview": { id: "alibaba/qwen3.6-max-preview", name: "Qwen3.6 Max Preview", description: "Flagship Qwen model for complex reasoning, coding, and agentic workflows", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2026-04-20", last_updated: "2026-04-20", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 65536 } }, "alibaba/qwen3.8-max": { id: "alibaba/qwen3.8-max", name: "Qwen3.8 Max", description: "2.4-trillion-parameter MoE flagship for coding, professional work, multimodal understanding, and long-horizon agentic workflows", family: "qwen", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-08-03", last_updated: "2026-08-03", modalities: { input: ["text", "image", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 131072 } }, "alibaba/qwen3.7-plus": { id: "alibaba/qwen3.7-plus", name: "Qwen3.7 Plus", description: "Multimodal Qwen workhorse for long-context agents, visual inputs, and coding", family: "qwen", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2026-06-02", last_updated: "2026-06-02", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 64e3 } }, "alibaba/qwq-32b": { id: "alibaba/qwq-32b", name: "QwQ 32B", description: "Open reasoning model from the Qwen team for math, coding, and step-by-step problem solving", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2025-03-05", last_updated: "2025-03-05", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/QwQ-32B" }] }, "alibaba/qwen3.7-flash": { id: "alibaba/qwen3.7-flash", name: "Qwen3.7 Flash", description: "Lightweight multimodal Qwen model for high-throughput text, image, and video tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-07-15", last_updated: "2026-07-15", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, input: 991e3, output: 65536 } }, "alibaba/qwen3-next-80b-a3b-thinking": { id: "alibaba/qwen3-next-80b-a3b-thinking", name: "Qwen3-Next 80B-A3B (Thinking)", description: "Efficient Qwen thinking model for local reasoning, math, and coding agents", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-09", last_updated: "2025-09", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking" }] }, "alibaba/qwen3-coder-480b-a35b-instruct": { id: "alibaba/qwen3-coder-480b-a35b-instruct", name: "Qwen3-Coder 480B-A35B Instruct", description: "Open Qwen coding heavyweight for repository reasoning and agentic engineering", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-04", last_updated: "2025-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct" }], benchmarks: [{ name: "SWE-Bench Pro", score: 38.7, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "alibaba/qwen2.5-coder-32b-instruct": { id: "alibaba/qwen2.5-coder-32b-instruct", name: "Qwen2.5-Coder-32B-Instruct", description: "Open coding-focused Qwen model for code generation, repair, and repository reasoning", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2024-11-12", last_updated: "2024-11-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct" }] }, "alibaba/qwen-flash": { id: "alibaba/qwen-flash", name: "Qwen Flash", description: "Efficient Qwen model for fast chat, extraction, and high-volume workloads", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2025-07-28", last_updated: "2025-07-28", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 32768 } }, "alibaba/qwen3-max": { id: "alibaba/qwen3-max", name: "Qwen3 Max", description: "Flagship Qwen3 model for coding agents, complex reasoning, and tool use", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-09-23", last_updated: "2025-09-23", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 65536 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 26.4, metric: "index", source: "https://openrouter.ai/qwen/qwen3-max/benchmarks", date: "2026-05-30" }, { name: "SciCode", score: 38.3, metric: "percent correct", source: "https://openrouter.ai/qwen/qwen3-max/benchmarks", date: "2026-05-30" }, { name: "Terminal-Bench Hard", score: 20.5, metric: "success rate", source: "https://openrouter.ai/qwen/qwen3-max/benchmarks", date: "2026-05-30" }] }, "alibaba/qwen3.6-35b-a3b": { id: "alibaba/qwen3.6-35b-a3b", name: "Qwen3.6 35B-A3B", description: "Open multimodal Qwen MoE for local agents that need vision, audio, and code", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-17", last_updated: "2026-04-17", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.6-35B-A3B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 73.4, metric: "resolved", source: "https://huggingface.co/Qwen/Qwen3.6-35B-A3B" }] }, "alibaba/qwen3-235b-a22b-instruct-2507": { id: "alibaba/qwen3-235b-a22b-instruct-2507", name: "Qwen3 235B-A22B Instruct 2507", description: "Updated large open Qwen3 MoE instruct model for multilingual chat, coding, and tool use", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2025-07-21", last_updated: "2025-07-21", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 16384 }, license: "Apache 2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507" }] }, "alibaba/qwen3.6-27b": { id: "alibaba/qwen3.6-27b", name: "Qwen3.6 27B", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-22", last_updated: "2026-04-22", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.6-27B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 77.2, metric: "resolved", source: "https://huggingface.co/Qwen/Qwen3.6-27B" }] }, "alibaba/qwen3-next-80b-a3b-instruct": { id: "alibaba/qwen3-next-80b-a3b-instruct", name: "Qwen3-Next 80B-A3B Instruct", description: "Qwen instruction model for multilingual chat, reasoning, and tool use", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-09", last_updated: "2025-09", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct" }] }, "alibaba/qwen3.8-27b": { id: "alibaba/qwen3.8-27b", name: "Qwen3.8 27B", description: "Dense 27B vision-language model for coding, agent tasks, and image and video understanding", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-14", last_updated: "2026-08-14", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.8-27B" }], benchmarks: [{ name: "SWE-bench Pro", score: 61.7, metric: "resolved", source: "https://huggingface.co/Qwen/Qwen3.8-27B" }] }, "alibaba/qwen3.6-plus": { id: "alibaba/qwen3.6-plus", name: "Qwen3.6 Plus", description: "Earlier Qwen multimodal workhorse for million-token agent and document tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2026-04-02", last_updated: "2026-04-02", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "alibaba/qwen3.5-9b": { id: "alibaba/qwen3.5-9b", name: "Qwen3.5 9B", description: "Qwen instruction model for multilingual chat, reasoning, and tool use", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-23", last_updated: "2026-02-23", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.5-9B" }] }, "alibaba/qwen3.5-27b": { id: "alibaba/qwen3.5-27b", name: "Qwen3.5 27B", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-23", last_updated: "2026-02-23", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.5-27B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 72.4, metric: "resolved", source: "https://huggingface.co/Qwen/Qwen3.5-27B" }] }, "alibaba/qwq-plus": { id: "alibaba/qwq-plus", name: "QwQ Plus", description: "Qwen reasoning model for deliberate problem solving, math, and coding", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2025-03-05", last_updated: "2025-03-05", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 8192 } }, "alibaba/qwen3.5-122b-a10b": { id: "alibaba/qwen3.5-122b-a10b", name: "Qwen3.5 122B-A10B", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-23", last_updated: "2026-02-23", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.5-122B-A10B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 72, metric: "resolved", source: "https://huggingface.co/Qwen/Qwen3.5-122B-A10B" }] }, "alibaba/qwen3-coder-plus": { id: "alibaba/qwen3-coder-plus", name: "Qwen3 Coder Plus", description: "Hosted Qwen coder for software agents, repo edits, and long-context code", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-07-23", last_updated: "2025-07-23", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "alibaba/qwen3-vl-235b-a22b-instruct": { id: "alibaba/qwen3-vl-235b-a22b-instruct", name: "Qwen3 VL 235B A22B Instruct", description: "Qwen vision-language instruct model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-09-23", last_updated: "2025-09-23", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Instruct" }] }, "alibaba/qwen-vl-plus": { id: "alibaba/qwen-vl-plus", name: "Qwen-VL Plus", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2024-01-25", last_updated: "2025-08-15", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 8192 } }, "alibaba/qwen3.6-flash": { id: "alibaba/qwen3.6-flash", name: "Qwen3.6 Flash", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen3.6", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-27", last_updated: "2026-04-27", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "alibaba/qwen3.8-2.4t-a95b": { id: "alibaba/qwen3.8-2.4t-a95b", name: "Qwen3.8 2.4T A95B", description: "Open-weight sparse MoE (2.4T total, 95B active), the open-weight twin of Qwen3.8 Max for coding, research, complex reasoning, and agentic workflows", family: "qwen", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-12", last_updated: "2026-08-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 131072 }, license: "qwen3.8-max", weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" }] }, "sakana/fugu": { id: "sakana/fugu", name: "Fugu", description: "Multi-agent model for routing expert agents across complex analytical tasks", family: "fugu", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, release_date: "2026-06-15", last_updated: "2026-06-15", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 1e6 }, links: [{ label: "Official model catalog", url: "https://raw.githubusercontent.com/SakanaAI/fugu/refs/heads/main/configs/files/fugu.json", type: "docs" }], benchmarks: [{ name: "SWE Bench Pro", score: 59, source: "https://console.sakana.ai/models" }, { name: "Terminal Bench 2.1", score: 80.2, source: "https://console.sakana.ai/models" }, { name: "LiveCodeBench", score: 92.9, source: "https://console.sakana.ai/models" }, { name: "LiveCodeBench Pro", score: 87.8, source: "https://console.sakana.ai/models" }, { name: "Humanity\u2019s Last Exam", score: 47.2, source: "https://console.sakana.ai/models" }, { name: "CharXiv Reasoning", score: 85.1, source: "https://console.sakana.ai/models" }, { name: "GPQA Diamond", score: 95.5, source: "https://console.sakana.ai/models" }, { name: "SciCode", score: 60.1, source: "https://console.sakana.ai/models" }, { name: "\u03C43 Banking", score: 21.7, source: "https://console.sakana.ai/models" }, { name: "Long Context Reasoning", score: 74.7, source: "https://console.sakana.ai/models" }, { name: "MRCRv2", score: 86.6, source: "https://console.sakana.ai/models" }, { name: "CTI-REALM", score: 67.5, source: "https://console.sakana.ai/models" }] }, "sakana/sakana-namazu": { id: "sakana/sakana-namazu", name: "Sakana Namazu", description: "Japanese-specialized reasoning model based on Kimi K2.6 and tuned for Japanese language, culture, and business workflows", family: "sakana-namazu", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-03", last_updated: "2026-08-03", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 65536 }, links: [{ label: "Official product page", url: "https://sakana.ai/namazu/", type: "announcement" }, { label: "Official model documentation", url: "https://console.sakana.ai/models?model=sakana-namazu", type: "docs" }], benchmarks: [{ name: "AIME26", score: 96.67, source: "https://console.sakana.ai/models?model=sakana-namazu" }, { name: "MMLU-Pro", score: 90.33, source: "https://console.sakana.ai/models?model=sakana-namazu" }, { name: "LiveCodeBench v6", score: 90.33, source: "https://console.sakana.ai/models?model=sakana-namazu" }, { name: "JFBench", score: 37.4, source: "https://console.sakana.ai/models?model=sakana-namazu" }, { name: "Translation", score: 52.2, source: "https://console.sakana.ai/models?model=sakana-namazu" }, { name: "FairPoliticsQA", score: 56.3, source: "https://console.sakana.ai/models?model=sakana-namazu" }] }, "sakana/fugu-ultra": { id: "sakana/fugu-ultra", name: "Fugu Ultra", description: "Quality-first multi-agent model for hard research, analysis, and competitions", family: "fugu", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, release_date: "2026-06-15", last_updated: "2026-06-15", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 1e6 }, links: [{ label: "Official model catalog", url: "https://raw.githubusercontent.com/SakanaAI/fugu/refs/heads/main/configs/files/fugu.json", type: "docs" }], benchmarks: [{ name: "SWE Bench Pro", score: 73.7, source: "https://console.sakana.ai/models" }, { name: "Terminal Bench 2.1", score: 82.1, source: "https://console.sakana.ai/models" }, { name: "LiveCodeBench", score: 93.2, source: "https://console.sakana.ai/models" }, { name: "LiveCodeBench Pro", score: 90.8, source: "https://console.sakana.ai/models" }, { name: "Humanity\u2019s Last Exam", score: 50, source: "https://console.sakana.ai/models" }, { name: "CharXiv Reasoning", score: 86.6, source: "https://console.sakana.ai/models" }, { name: "GPQA Diamond", score: 95.5, source: "https://console.sakana.ai/models" }, { name: "SciCode", score: 58.7, source: "https://console.sakana.ai/models" }, { name: "\u03C43 Banking", score: 20.6, source: "https://console.sakana.ai/models" }, { name: "Long Context Reasoning", score: 73.3, source: "https://console.sakana.ai/models" }, { name: "MRCRv2", score: 93.6, source: "https://console.sakana.ai/models" }, { name: "CTI-REALM", score: 69.4, source: "https://console.sakana.ai/models" }] }, "perplexity/sonar-deep-research": { id: "perplexity/sonar-deep-research", name: "Sonar Deep Research", description: "Sonar search model for autonomous research and citation-backed long-form reports", family: "sonar", attachment: false, reasoning: true, tool_call: false, temperature: false, knowledge: "2025-01", release_date: "2025-02-01", last_updated: "2025-09-01", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 32768 } }, "perplexity/sonar": { id: "perplexity/sonar", name: "Sonar", description: "Fast web-grounded Sonar for current answers, citations, and lightweight retrieval", family: "sonar", attachment: false, reasoning: false, tool_call: false, temperature: true, knowledge: "2025-09-01", release_date: "2024-01-01", last_updated: "2025-09-01", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 4096 }, benchmarks: [{ name: "SciCode", score: 22.9, metric: "percent correct", source: "https://openrouter.ai/perplexity/sonar/benchmarks", date: "2026-03-11" }] }, "perplexity/sonar-reasoning-pro": { id: "perplexity/sonar-reasoning-pro", name: "Sonar Reasoning Pro", description: "Web-grounded Sonar for multi-step research questions that need cited reasoning", family: "sonar-reasoning", attachment: true, reasoning: true, tool_call: false, temperature: true, knowledge: "2025-09-01", release_date: "2024-01-01", last_updated: "2025-09-01", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 4096 } }, "perplexity/sonar-pro": { id: "perplexity/sonar-pro", name: "Sonar Pro", description: "Deeper Sonar search model with broader retrieval and stronger synthesis", family: "sonar-pro", attachment: true, reasoning: false, tool_call: false, temperature: true, knowledge: "2025-09-01", release_date: "2024-01-01", last_updated: "2025-09-01", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 8192 }, benchmarks: [{ name: "SciCode", score: 22.6, metric: "percent correct", source: "https://openrouter.ai/perplexity/sonar-pro/benchmarks", date: "2026-03-11" }] }, "zhipuai/glm-4.6": { id: "zhipuai/glm-4.6", name: "GLM-4.6", description: "Late GLM-4 workhorse for coding agents, reasoning, and structured tasks", family: "glm", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-09-30", last_updated: "2025-09-30", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.6" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 29.5, metric: "index", source: "https://openrouter.ai/z-ai/glm-4.6/benchmarks", date: "2026-05-22" }, { name: "SciCode", score: 38.4, metric: "percent correct", source: "https://openrouter.ai/z-ai/glm-4.6/benchmarks", date: "2026-05-22" }, { name: "Terminal-Bench Hard", score: 25, metric: "success rate", source: "https://openrouter.ai/z-ai/glm-4.6/benchmarks", date: "2026-05-22" }, { name: "SWE-Bench Pro", score: 9.67, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "zhipuai/glm-4.5-flash": { id: "zhipuai/glm-4.5-flash", name: "GLM-4.5-Flash", description: "Efficient GLM model for fast reasoning, coding, and agent workflows", family: "glm-flash", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-07-28", last_updated: "2025-07-28", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 98304 } }, "zhipuai/glm-5v-turbo": { id: "zhipuai/glm-5v-turbo", name: "GLM-5V-Turbo", description: "Fast GLM vision model for screenshots, documents, and multimodal agent tasks", family: "glm", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-04-01", last_updated: "2026-04-01", modalities: { input: ["text", "image", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 131072 } }, "zhipuai/glm-4.7": { id: "zhipuai/glm-4.7", name: "GLM-4.7", description: "Mature GLM model for dependable coding, reasoning, and structured agent tasks", family: "glm", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-12-22", last_updated: "2025-12-22", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.7" }], benchmarks: [{ name: "SWE-Bench Verified", score: 73.8, metric: "resolved", source: "https://huggingface.co/zai-org/GLM-4.7" }, { name: "Terminal Bench 2.0", score: 33.4, metric: "score", source: "https://huggingface.co/zai-org/GLM-4.7" }] }, "zhipuai/glm-4.5v": { id: "zhipuai/glm-4.5v", name: "GLM-4.5V", description: "GLM vision model for visual reasoning, documents, and multimodal agents", family: "glm", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-08-11", last_updated: "2025-08-11", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 64e3, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.5V" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 10.9, metric: "index", source: "https://openrouter.ai/z-ai/glm-4.5v/benchmarks", date: "2026-04-29" }, { name: "SciCode", score: 22.1, metric: "percent correct", source: "https://openrouter.ai/z-ai/glm-4.5v/benchmarks", date: "2026-04-29" }, { name: "Terminal-Bench Hard", score: 5.3, metric: "success rate", source: "https://openrouter.ai/z-ai/glm-4.5v/benchmarks", date: "2026-04-29" }] }, "zhipuai/glm-4.6v-flash": { id: "zhipuai/glm-4.6v-flash", name: "GLM-4.6V-Flash", description: "Lightweight GLM vision model for visual reasoning, documents, and multimodal agents", family: "glm", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2025-12-08", last_updated: "2025-12-08", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.6V-Flash" }] }, "zhipuai/glm-4.7-flashx": { id: "zhipuai/glm-4.7-flashx", name: "GLM-4.7-FlashX", description: "Efficient GLM model for fast reasoning, coding, and agent workflows", family: "glm-flash", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2026-01-19", last_updated: "2026-01-19", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 2e5, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.7-Flash" }] }, "zhipuai/glm-5": { id: "zhipuai/glm-5", name: "GLM-5", description: "General GLM flagship for coding, analysis, and tool-heavy engineering workflows", family: "glm", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-02-12", last_updated: "2026-02-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-5" }], benchmarks: [{ name: "SWE-Bench Verified", score: 72.8, metric: "resolved", source: "https://www.swebench.com/" }, { name: "SWE-Atlas Codebase QnA", score: 20.5, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 24.24, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 28.74, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }] }, "zhipuai/glm-5-turbo": { id: "zhipuai/glm-5-turbo", name: "GLM-5-Turbo", description: "Faster GLM-5 lane for coding agents that need lower latency", family: "glm", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-03-16", last_updated: "2026-03-16", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 131072 } }, "zhipuai/glm-5.3": { id: "zhipuai/glm-5.3", name: "GLM-5.3", description: "Flagship GLM model for long-horizon coding, agents, and complex project delivery", family: "glm", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-14", last_updated: "2026-08-14", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 131072 } }, "zhipuai/glm-5.2": { id: "zhipuai/glm-5.2", name: "GLM-5.2", description: "Open flagship GLM for long-horizon coding agents and million-token context work", family: "glm", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-06-13", last_updated: "2026-06-13", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-5.2" }], benchmarks: [{ name: "SWE-Bench Pro", score: 62.1, metric: "resolve rate", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "Terminal-Bench", score: 82.7, metric: "success rate", harness: "Claude Code", version: "2.1", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "FrontierSWE", score: 74.4, metric: "dominance", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "Humanity's Last Exam", score: 40.5, metric: "accuracy", dataset: "text-only subset", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "Humanity's Last Exam", score: 54.7, metric: "accuracy", variant: "with tools", dataset: "text-only subset", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "CritPt", score: 20.9, metric: "accuracy", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "AIME", score: 99.2, metric: "accuracy", version: "2026", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "HMMT", score: 94.4, metric: "accuracy", version: "November 2025", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "HMMT", score: 92.5, metric: "accuracy", version: "February 2026", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "IMOAnswerBench", score: 91, metric: "accuracy", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "GPQA Diamond", score: 91.2, metric: "accuracy", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "NL2Repo", score: 48.9, metric: "resolve rate", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "DeepSWE", score: 46.2, metric: "resolve rate", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "Program Bench", score: 63.7, metric: "score", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "Terminal-Bench", score: 81, metric: "success rate", harness: "Terminus 2", version: "2.1", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "PostTrainBench", score: 34.3, metric: "score", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "SWE Marathon", score: 13, metric: "resolve rate", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "MCP Atlas", score: 76.8, metric: "score", dataset: "public subset", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "Tool-Decathlon", score: 48.2, metric: "score", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }] }, "zhipuai/glm-4.6v": { id: "zhipuai/glm-4.6v", name: "GLM-4.6V", description: "GLM vision model for visual reasoning, documents, and multimodal agents", family: "glm", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-12-08", last_updated: "2025-12-08", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.6V" }] }, "zhipuai/glm-4.5-air": { id: "zhipuai/glm-4.5-air", name: "GLM-4.5-Air", description: "Lighter GLM-4.5 variant for fast coding assistance and cheaper agents", family: "glm-air", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-07-28", last_updated: "2025-07-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 98304 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.5-Air" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 23.8, metric: "index", source: "https://openrouter.ai/z-ai/glm-4.5-air/benchmarks", date: "2026-05-30" }, { name: "SciCode", score: 30.6, metric: "percent correct", source: "https://openrouter.ai/z-ai/glm-4.5-air/benchmarks", date: "2026-05-30" }, { name: "Terminal-Bench Hard", score: 20.5, metric: "success rate", source: "https://openrouter.ai/z-ai/glm-4.5-air/benchmarks", date: "2026-05-30" }] }, "zhipuai/glm-5.1": { id: "zhipuai/glm-5.1", name: "GLM-5.1", description: "Strong GLM coding model for agentic engineering, terminals, and repository generation", family: "glm", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-07", last_updated: "2026-04-07", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 2e5, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-5.1" }], benchmarks: [{ name: "Artificial Analysis Coding Agent Index", score: 52.7, metric: "average pass@1", harness: "Claude Code", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 73.2, metric: "pass@1", harness: "Claude Code", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 19.8, metric: "pass@1", harness: "Claude Code", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 65.1, metric: "pass@1", harness: "Claude Code", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }] }, "zhipuai/glm-4.7-flash": { id: "zhipuai/glm-4.7-flash", name: "GLM-4.7-Flash", description: "Budget GLM lane for fast coding help, routing, and everyday automation", family: "glm-flash", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2026-01-19", last_updated: "2026-01-19", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 2e5, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.7-Flash" }], benchmarks: [{ name: "SWE-Bench Verified", score: 59.2, metric: "resolved", source: "https://huggingface.co/zai-org/GLM-4.7-Flash" }] }, "zhipuai/glm-4.5": { id: "zhipuai/glm-4.5", name: "GLM-4.5", description: "Hybrid-reasoning GLM release that made the 4.5 line broadly useful", family: "glm", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-07-28", last_updated: "2025-07-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 98304 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.5" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 26.3, metric: "index", source: "https://openrouter.ai/z-ai/glm-4.5/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 34.8, metric: "percent correct", source: "https://openrouter.ai/z-ai/glm-4.5/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 22, metric: "success rate", source: "https://openrouter.ai/z-ai/glm-4.5/benchmarks", date: "2026-03-11" }] }, "ibm/granite-4-h-micro": { id: "ibm/granite-4-h-micro", name: "Granite-4.0-H-Micro", description: "Compact open-weight hybrid Granite model for lightweight enterprise chat and tool calling", family: "granite", attachment: false, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2025-10-02", last_updated: "2025-10-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/ibm-granite/granite-4.0-h-micro" }] }, "ibm/granite-4-h-small": { id: "ibm/granite-4-h-small", name: "Granite-4.0-H-Small", description: "Open-weight hybrid model for enterprise chat, coding, retrieval-augmented generation, and tool-calling workloads", family: "granite", attachment: false, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2025-10-02", last_updated: "2025-10-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/ibm-granite/granite-4.0-h-small" }] }, "thinkingmachines/inkling-small": { id: "thinkingmachines/inkling-small", name: "Inkling Small", description: "Multimodal MoE reasoning model (276B total, 12B active) for text, image, and audio", family: "ling", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-07-30", last_updated: "2026-07-30", modalities: { input: ["text", "image", "audio"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 1048576 }, license: "Apache-2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/thinkingmachines/Inkling-Small" }] }, "thinkingmachines/inkling": { id: "thinkingmachines/inkling", name: "Inkling", description: "Multimodal MoE reasoning model (975B total, 41B active) for text, image, and audio", family: "ling", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-07-15", last_updated: "2026-07-15", modalities: { input: ["text", "image", "audio"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 1048576 }, license: "Apache-2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/thinkingmachines/Inkling" }] }, "mistral/magistral-small-2506": { id: "mistral/magistral-small-2506", name: "Magistral Small", description: "Open Mistral reasoning model for transparent step-by-step problem solving", family: "magistral", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-06-10", last_updated: "2025-06-10", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, license: "Apache 2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Magistral-Small-2506" }] }, "mistral/mistral-small-latest": { id: "mistral/mistral-small-latest", name: "Mistral Small (latest)", description: "Efficient Mistral model for fast chat, extraction, and production assistants", family: "mistral-small", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-06", release_date: "2026-03-16", last_updated: "2026-03-16", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 256e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603" }] }, "mistral/devstral-medium-2507": { id: "mistral/devstral-medium-2507", name: "Devstral Medium", description: "Mistral coding agent model for repository tasks and software engineering workflows", family: "devstral", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-05", release_date: "2025-07-10", last_updated: "2025-07-10", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Verified", score: 61.6, metric: "resolved", source: "https://mistral.ai/news/devstral-2507", date: "2025-07-10" }] }, "mistral/mistral-small-3-1-24b-instruct-2503": { id: "mistral/mistral-small-3-1-24b-instruct-2503", name: "Mistral Small 3.1 24B", description: "Efficient multimodal model for instruction following, coding, reasoning, and function calling", family: "mistral-small", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-06", release_date: "2025-03-17", last_updated: "2025-03-17", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503" }] }, "mistral/mistral-large-2411": { id: "mistral/mistral-large-2411", name: "Mistral Large 2.1", description: "Flagship Mistral model for advanced reasoning, coding, and multilingual work", family: "mistral-large", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-11", release_date: "2024-11-18", last_updated: "2024-11-18", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 13.8, metric: "index", source: "https://openrouter.ai/mistralai/mistral-large-2407/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 29.2, metric: "percent correct", source: "https://openrouter.ai/mistralai/mistral-large-2407/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 6.1, metric: "success rate", source: "https://openrouter.ai/mistralai/mistral-large-2407/benchmarks", date: "2026-03-11" }] }, "mistral/mistral-nemo": { id: "mistral/mistral-nemo", name: "Mistral Nemo", description: "Efficient Mistral-NVIDIA open model for multilingual chat and local deployment", family: "mistral-nemo", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-07", release_date: "2024-07-01", last_updated: "2024-07-01", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407" }] }, "mistral/mistral-large-latest": { id: "mistral/mistral-large-latest", name: "Mistral Large (latest)", description: "Flagship Mistral model for advanced reasoning, coding, and multilingual work", family: "mistral-large", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-11", release_date: "2024-11-01", last_updated: "2025-12-02", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512" }] }, "mistral/ministral-8b-instruct-2410": { id: "mistral/ministral-8b-instruct-2410", name: "Ministral 8B Instruct", description: "Efficient open Mistral edge model for on-device chat and function calling", family: "ministral", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2024-10-16", last_updated: "2024-10-16", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, license: "Mistral Research License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Ministral-8B-Instruct-2410" }] }, "mistral/devstral-small-2507": { id: "mistral/devstral-small-2507", name: "Devstral Small", description: "Mistral coding agent model for repository tasks and software engineering workflows", family: "devstral", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-05", release_date: "2025-07-10", last_updated: "2025-07-10", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Devstral-Small-2507" }], benchmarks: [{ name: "SWE-Bench Verified", score: 53.6, metric: "resolved", source: "https://mistral.ai/news/devstral-2507", date: "2025-07-10" }] }, "mistral/devstral-2512": { id: "mistral/devstral-2512", name: "Devstral 2", description: "Mistral's coding-agent model for repository work, terminal tasks, and software fixes", family: "devstral", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-12", release_date: "2025-12-09", last_updated: "2025-12-09", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 23.7, metric: "index", source: "https://openrouter.ai/mistralai/devstral-2512/benchmarks", date: "2026-05-31" }, { name: "SciCode", score: 33.1, metric: "percent correct", source: "https://openrouter.ai/mistralai/devstral-2512/benchmarks", date: "2026-05-31" }, { name: "Terminal-Bench Hard", score: 18.9, metric: "success rate", source: "https://openrouter.ai/mistralai/devstral-2512/benchmarks", date: "2026-05-31" }] }, "mistral/mistral-small-2603": { id: "mistral/mistral-small-2603", name: "Mistral Small 4", description: "Fast Mistral production model for chat, extraction, and cost-sensitive agents", family: "mistral-small", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-06", release_date: "2026-03-16", last_updated: "2026-03-16", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 256e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 24.3, metric: "index", source: "https://openrouter.ai/mistralai/mistral-small-2603/benchmarks", date: "2026-06-01" }, { name: "SciCode", score: 38, metric: "percent correct", source: "https://openrouter.ai/mistralai/mistral-small-2603/benchmarks", date: "2026-06-01" }, { name: "Terminal-Bench Hard", score: 17.4, metric: "success rate", source: "https://openrouter.ai/mistralai/mistral-small-2603/benchmarks", date: "2026-06-01" }] }, "mistral/mistral-small-2506": { id: "mistral/mistral-small-2506", name: "Mistral Small 3.2", description: "Efficient Mistral model for fast chat, extraction, and production assistants", family: "mistral-small", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-03", release_date: "2025-06-20", last_updated: "2025-06-20", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506" }] }, "mistral/pixtral-large-latest": { id: "mistral/pixtral-large-latest", name: "Pixtral Large (latest)", description: "Mistral's larger vision model for document-heavy image understanding and chat", family: "pixtral", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-11", release_date: "2024-11-01", last_updated: "2024-11-04", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Pixtral-Large-Instruct-2411" }] }, "mistral/mistral-large-2512": { id: "mistral/mistral-large-2512", name: "Mistral Large 3", description: "Mistral's largest general model for enterprise agents, coding, and multilingual reasoning", family: "mistral-large", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-11", release_date: "2024-11-01", last_updated: "2025-12-02", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 22.7, metric: "index", source: "https://openrouter.ai/mistralai/mistral-large-2512/benchmarks", date: "2026-06-04" }, { name: "SciCode", score: 36.2, metric: "percent correct", source: "https://openrouter.ai/mistralai/mistral-large-2512/benchmarks", date: "2026-06-04" }, { name: "Terminal-Bench Hard", score: 15.9, metric: "success rate", source: "https://openrouter.ai/mistralai/mistral-large-2512/benchmarks", date: "2026-06-04" }] }, "mistral/mistral-medium-2604": { id: "mistral/mistral-medium-2604", name: "Mistral Medium 3.5", description: "Balanced Mistral model for enterprise assistants, multilingual work, and tools", family: "mistral-medium", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-29", last_updated: "2026-04-29", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 77.6, metric: "resolved", source: "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B" }, { name: "\u03C4\xB3-Telecom", score: 91.4, metric: "accuracy", variant: "public preview", source: "https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5/", date: "2026-05-22" }] }, "mistral/mistral-medium-2505": { id: "mistral/mistral-medium-2505", name: "Mistral Medium 3", description: "Mistral model for multilingual chat, reasoning, and tool-assisted workflows", family: "mistral-medium", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-05", release_date: "2025-05-07", last_updated: "2025-05-07", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 131072 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 13.6, metric: "index", source: "https://openrouter.ai/mistralai/mistral-medium-3/benchmarks", date: "2026-05-30" }, { name: "SciCode", score: 33.1, metric: "percent correct", source: "https://openrouter.ai/mistralai/mistral-medium-3/benchmarks", date: "2026-05-30" }, { name: "Terminal-Bench Hard", score: 3.8, metric: "success rate", source: "https://openrouter.ai/mistralai/mistral-medium-3/benchmarks", date: "2026-05-30" }] }, "mistral/devstral-medium-latest": { id: "mistral/devstral-medium-latest", name: "Devstral 2 (latest)", description: "Mistral coding agent model for repository tasks and software engineering workflows", family: "devstral", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-12", release_date: "2025-12-02", last_updated: "2025-12-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512" }] }, "mistral/pixtral-12b": { id: "mistral/pixtral-12b", name: "Pixtral 12B", description: "Mistral vision-language model for image understanding and multimodal chat", family: "pixtral", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-09", release_date: "2024-09-01", last_updated: "2024-09-01", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Pixtral-12B-2409" }] }, "mistral/voxtral-small-latest": { id: "mistral/voxtral-small-latest", name: "Voxtral Small (latest)", description: "Instruct model with native audio input for speech understanding and tool use", family: "voxtral", attachment: true, reasoning: false, tool_call: true, temperature: true, release_date: "2025-07-15", last_updated: "2025-07-15", modalities: { input: ["text", "audio"], output: ["text"] }, open_weights: true, limit: { context: 32e3, output: 32e3 } }, "mistral/codestral-22b-v0.1": { id: "mistral/codestral-22b-v0.1", name: "Codestral-22B-v0.1", description: "Open Mistral code model for fill-in-the-middle and 80+ programming languages", family: "codestral", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2024-05-29", last_updated: "2024-05-29", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 32768, output: 8192 }, license: "Mistral AI Non-Production License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Codestral-22B-v0.1" }] }, "mistral/codestral-latest": { id: "mistral/codestral-latest", name: "Codestral (latest)", description: "Mistral code model for completions, refactors, and developer IDE workflows", family: "codestral", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-10", release_date: "2024-05-29", last_updated: "2025-01-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Codestral-22B-v0.1" }], benchmarks: [{ name: "Aider Polyglot", score: 11.1, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-01-13" }] }, "mistral/magistral-medium-latest": { id: "mistral/magistral-medium-latest", name: "Magistral Medium (latest)", description: "Mistral reasoning model for transparent analysis, math, and complex decisions", family: "magistral-medium", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-06", release_date: "2025-03-17", last_updated: "2025-03-20", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 } }, "mistral/mistral-medium-latest": { id: "mistral/mistral-medium-latest", name: "Mistral Medium (latest)", description: "Balanced Mistral model for enterprise assistants, multilingual work, and tools", family: "mistral-medium", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-29", last_updated: "2026-04-29", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 77.6, metric: "resolved", source: "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B" }] }, "upstage/solar-pro2": { id: "upstage/solar-pro2", name: "Solar Pro 2", description: "Flagship model for demanding analysis, coding, and production agent workflows", family: "solar-pro", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03", release_date: "2025-05-20", last_updated: "2025-05-20", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 65536, output: 8192 } }, "upstage/solar-pro4": { id: "upstage/solar-pro4", name: "Solar Pro 4", description: "Upstage's flagship model, specialized for agentic use", family: "solar-pro", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-02", release_date: "2026-08-06", last_updated: "2026-08-06", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 524288, output: 131072 } }, "upstage/solar-pro3": { id: "upstage/solar-pro3", name: "Solar Pro 3", description: "Flagship model for demanding analysis, coding, and production agent workflows", family: "solar-pro", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03", release_date: "2026-01", last_updated: "2026-01", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 8192 } }, "minimax/MiniMax-M2.7": { id: "minimax/MiniMax-M2.7", name: "MiniMax-M2.7", description: "Open MiniMax flagship for coding agents, office automation, and complex environments", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-03-18", last_updated: "2026-03-18", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M2.7" }], benchmarks: [{ name: "SWE-Bench Verified", score: 79.9, metric: "resolved", harness: "Claude Code", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "SWE-Bench Pro", score: 56.2, metric: "resolve rate", harness: "Claude Code", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "Terminal-Bench", score: 51.1, metric: "success rate", version: "2.1", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }] }, "minimax/MiniMax-M2.7-highspeed": { id: "minimax/MiniMax-M2.7-highspeed", name: "MiniMax-M2.7-highspeed", description: "Low-latency M2.7 variant for interactive coding plans and agent loops", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-03-18", last_updated: "2026-03-18", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M2.7" }] }, "minimax/MiniMax-M2.1": { id: "minimax/MiniMax-M2.1", name: "MiniMax-M2.1", description: "Earlier MiniMax agent model for practical coding and productivity tasks", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-12-23", last_updated: "2025-12-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M2.1" }], benchmarks: [{ name: "SWE-Bench Verified", score: 74, metric: "resolved", source: "https://huggingface.co/MiniMaxAI/MiniMax-M2.1" }, { name: "SWE-Bench Pro", score: 36.81, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "minimax/MiniMax-M2": { id: "minimax/MiniMax-M2", name: "MiniMax-M2", description: "Efficient open MiniMax model built for coding agents and tool-heavy workflows", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-10-27", last_updated: "2025-10-27", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 196608, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M2" }], benchmarks: [{ name: "SWE-Bench Verified", score: 69.4, metric: "resolved", source: "https://huggingface.co/MiniMaxAI/MiniMax-M2" }] }, "minimax/MiniMax-M2.5-highspeed": { id: "minimax/MiniMax-M2.5-highspeed", name: "MiniMax-M2.5-highspeed", description: "High-speed MiniMax model for low-latency coding and agent workflows", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-02-13", last_updated: "2026-02-13", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M2.5" }] }, "minimax/MiniMax-M3": { id: "minimax/MiniMax-M3", name: "MiniMax-M3", description: "MiniMax multimodal model for long-context coding, perception, and agent planning", family: "minimax", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-01", last_updated: "2026-06-01", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 512e3, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M3" }], benchmarks: [{ name: "SWE-Bench Verified", score: 80.5, metric: "resolved", harness: "Claude Code", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "SWE-Bench Pro", score: 59, metric: "resolve rate", harness: "Claude Code", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "Terminal-Bench", score: 66, metric: "success rate", version: "2.1", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "BrowseComp", score: 83.52, metric: "accuracy", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "MCP Atlas", score: 74.2, metric: "success rate", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "OSWorld-Verified", score: 70.06, metric: "success rate", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }] }, "minimax/MiniMax-M2-Her": { id: "minimax/MiniMax-M2-Her", name: "MiniMax-M2 Her", description: "MiniMax M2 variant tuned for conversational and character-driven agent interactions", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-01-23", last_updated: "2026-01-23", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 131e3 } }, "minimax/MiniMax-M2.5": { id: "minimax/MiniMax-M2.5", name: "MiniMax-M2.5", description: "Prior MiniMax coding model for agent workflows, office edits, and automation", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-02-12", last_updated: "2026-02-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M2.5" }], benchmarks: [{ name: "SWE-Bench Verified", score: 75.8, metric: "resolved", source: "https://www.swebench.com/" }, { name: "SWE-Atlas Codebase QnA", score: 10.3, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 19.52, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 18.6, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }] } } };
+var registry_snapshot_default = { fetchedAt: "2026-09-16T13:32:11.933Z", count: 400, models: { "swiss-ai/apertus-8b": { id: "swiss-ai/apertus-8b", name: "Apertus 8B", description: "Fully open 8B multilingual LLM supporting 1800+ languages with 65K context. Trained on compliant open data. Apache 2.0, EU AI Act compliant.", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-09", release_date: "2025-09-02", last_updated: "2025-09-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 65536, output: 8192 }, license: "Apache-2.0", links: [{ label: "Paper", url: "https://arxiv.org/abs/2509.14233", type: "paper" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/swiss-ai/Apertus-8B-Instruct-2509" }] }, "swiss-ai/apertus-70b": { id: "swiss-ai/apertus-70b", name: "Apertus 70B", description: "Fully open 70B multilingual LLM supporting 1800+ languages with 65K context. Trained on 15T tokens of compliant open data. Apache 2.0, EU AI Act compliant.", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-09", release_date: "2025-09-02", last_updated: "2025-09-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 65536, output: 8192 }, license: "Apache-2.0", links: [{ label: "Paper", url: "https://arxiv.org/abs/2509.14233", type: "paper" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509" }] }, "meituan/longcat-2.0": { id: "meituan/longcat-2.0", name: "LongCat-2.0", description: "Meituan LongCat-2.0, a reasoning model with tool calling and a 1M-token context window", family: "longcat", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-30", last_updated: "2026-06-30", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 131072 }, benchmarks: [{ name: "SWE-Bench Pro", score: 59.5, metric: "resolve rate", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }, { name: "SWE-Bench Multilingual", score: 77.3, metric: "resolve rate", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }, { name: "Terminal-Bench", score: 70.8, metric: "success rate", version: "2.1", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }, { name: "GPQA Diamond", score: 88.9, metric: "accuracy", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }, { name: "BrowseComp", score: 79.9, metric: "accuracy", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }, { name: "IFEval", score: 90, metric: "accuracy", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }, { name: "FORTE", score: 73.2, metric: "success rate", source: "https://github.com/meituan-longcat/longcat-2.0", date: "2026-06-30" }] }, "poolside/laguna-xs-2.1": { id: "poolside/laguna-xs-2.1", name: "Laguna XS 2.1", description: "Agentic coding model from Poolside in the XS size class for local deployment", family: "laguna", attachment: false, reasoning: true, tool_call: true, structured_output: false, temperature: true, release_date: "2026-07-02", last_updated: "2026-07-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 32768 }, benchmarks: [{ name: "SWE-Bench Verified", score: 70.9, metric: "resolved", harness: "Harbor", source: "https://poolside.ai/blog/introducing-laguna-xs-2-1", date: "2026-07-02" }, { name: "SWE-Bench Multilingual", score: 63.1, metric: "resolve rate", harness: "Harbor", source: "https://poolside.ai/blog/introducing-laguna-xs-2-1", date: "2026-07-02" }, { name: "SWE-Bench Pro", score: 47.6, metric: "resolve rate", harness: "Harbor", source: "https://poolside.ai/blog/introducing-laguna-xs-2-1", date: "2026-07-02" }, { name: "Terminal-Bench", score: 37.5, metric: "success rate", harness: "Harbor", version: "2.0", source: "https://poolside.ai/blog/introducing-laguna-xs-2-1", date: "2026-07-02" }] }, "poolside/laguna-m.1": { id: "poolside/laguna-m.1", name: "Laguna M.1", description: "Poolside's open-weight model for agentic coding and long-horizon work", family: "laguna", attachment: false, reasoning: true, tool_call: true, structured_output: false, temperature: true, release_date: "2026-04-28", last_updated: "2026-06-13", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 32768 } }, "poolside/laguna-xs.2": { id: "poolside/laguna-xs.2", name: "Laguna XS.2", description: "Agentic coding model from Poolside in the XS size class for local deployment", family: "laguna", attachment: false, reasoning: true, tool_call: true, structured_output: false, temperature: true, release_date: "2026-04-28", last_updated: "2026-06-13", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 32768 } }, "poolside/laguna-s-2.1": { id: "poolside/laguna-s-2.1", name: "Laguna S 2.1", description: "Agentic coding model from Poolside in the XS size class for local deployment", family: "laguna", attachment: false, reasoning: true, tool_call: true, structured_output: false, temperature: true, release_date: "2026-07-21", last_updated: "2026-07-21", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 32768 } }, "deepreinforce/ornith-1.0-397b": { id: "deepreinforce/ornith-1.0-397b", name: "Ornith 1.0 397B", description: "Large coding-reasoning model for agentic software tasks and RL search", family: "ornith", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-25", last_updated: "2026-06-25", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144 }, license: "MIT", links: [{ label: "Model card", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B", type: "model_card" }, { label: "Announcement", url: "https://deep-reinforce.com/ornith_1_0.html", type: "announcement" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { label: "Hugging Face (FP8)", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B-FP8", quantization: "fp8" }], benchmarks: [{ name: "SWE-Bench Verified", score: 82.4, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { name: "SWE-Bench Pro", score: 62.2, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { name: "SWE-Bench Multilingual", score: 78.9, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { name: "Terminal-Bench 2.1", score: 77.5, metric: "percent", variant: "Terminus-2", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { name: "Terminal-Bench 2.1", score: 78.2, metric: "percent", variant: "Claude Code", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { name: "NL2Repo", score: 48.2, metric: "percent", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }, { name: "Claw-eval", score: 77.1, metric: "percent", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B" }] }, "deepreinforce/ornith-1.0-35b": { id: "deepreinforce/ornith-1.0-35b", name: "Ornith 1.0 35B", description: "Large coding-reasoning model for agentic software tasks and RL search", family: "ornith", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-25", last_updated: "2026-06-25", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144 }, license: "MIT", links: [{ label: "Model card", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B", type: "model_card" }, { label: "Announcement", url: "https://deep-reinforce.com/ornith_1_0.html", type: "announcement" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 75.6, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }, { name: "SWE-Bench Pro", score: 50.4, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }, { name: "SWE-Bench Multilingual", score: 69.3, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }, { name: "Terminal-Bench 2.1", score: 64.2, metric: "percent", variant: "Terminus-2", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }, { name: "Terminal-Bench 2.1", score: 62.8, metric: "percent", variant: "Claude Code", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }, { name: "NL2Repo", score: 34.6, metric: "percent", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }, { name: "Claw-eval", score: 69.8, metric: "percent", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B" }] }, "deepreinforce/ornith-1.0-9b": { id: "deepreinforce/ornith-1.0-9b", name: "Ornith 1.0 9B", description: "Open coding-reasoning model for repository tasks and self-improving agents", family: "ornith", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-25", last_updated: "2026-06-25", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144 }, license: "MIT", links: [{ label: "Model card", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B", type: "model_card" }, { label: "Announcement", url: "https://deep-reinforce.com/ornith_1_0.html", type: "announcement" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 69.4, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }, { name: "SWE-Bench Pro", score: 42.9, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }, { name: "SWE-Bench Multilingual", score: 52, metric: "percent resolved", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }, { name: "Terminal-Bench 2.1", score: 43.1, metric: "percent", variant: "Terminus-2", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }, { name: "Terminal-Bench 2.1", score: 40.6, metric: "percent", variant: "Claude Code", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }, { name: "NL2Repo", score: 27.2, metric: "percent", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }, { name: "Claw-eval", score: 63.1, metric: "percent", source: "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B" }] }, "deepreinforce/ornith-1.5-35b-a3b": { id: "deepreinforce/ornith-1.5-35b-a3b", name: "Ornith 1.5 35B A3B", description: "Mixture-of-experts coding-reasoning model for agentic software tasks, tool use, and image understanding", family: "ornith", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-18", last_updated: "2026-08-23", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144 }, license: "MIT", links: [{ label: "Model card", url: "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B", type: "model_card" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B" }] }, "deepreinforce/ornith-1.0-31b": { id: "deepreinforce/ornith-1.0-31b", name: "Ornith 1.0 31B", description: "Open coding-reasoning model for repository tasks and self-improving agents", family: "ornith", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-25", last_updated: "2026-06-25", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144 }, license: "MIT", links: [{ label: "Announcement", url: "https://deep-reinforce.com/ornith_1_0.html", type: "announcement" }] }, "stepfun/step-3.7-flash": { id: "stepfun/step-3.7-flash", name: "Step 3.7 Flash", description: "Newer StepFun flash model for faster agents, coding, and multimodal prompts", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2026-03-01", release_date: "2026-05-29", last_updated: "2026-05-29", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 256e3, input: 256e3, output: 256e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/stepfun-ai/Step-3.7-Flash" }], benchmarks: [{ name: "SWE-Bench Pro", score: 56.3, metric: "resolve rate", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "SWE-Bench Verified", score: 76.5, metric: "resolved", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "Terminal-Bench", score: 59.6, metric: "success rate", version: "2.1", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "Humanity's Last Exam", score: 47.2, metric: "accuracy", variant: "with tools", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "BrowseComp", score: 75.8, metric: "accuracy", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "Toolathlon", score: 49.5, metric: "success rate", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "GDPval", score: 45.8, metric: "wins or ties", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "ClawEval", score: 67.1, metric: "pass^3", version: "1.1", source: "https://static.stepfun.com/blog/step-3.7-flash/", date: "2026-05-29" }, { name: "Artificial Analysis Coding Index", score: 37.1, metric: "index", source: "https://openrouter.ai/stepfun/step-3.7-flash/benchmarks", date: "2026-06-15" }, { name: "SciCode", score: 40, metric: "percent correct", source: "https://openrouter.ai/stepfun/step-3.7-flash/benchmarks", date: "2026-06-15" }, { name: "Terminal-Bench Hard", score: 35.6, metric: "success rate", source: "https://openrouter.ai/stepfun/step-3.7-flash/benchmarks", date: "2026-06-15" }] }, "stepfun/step-3.5-flash": { id: "stepfun/step-3.5-flash", name: "Step 3.5 Flash", description: "StepFun flash lane for quick multimodal reasoning and coding assistance", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-01", release_date: "2026-01-29", last_updated: "2026-02-13", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, input: 256e3, output: 256e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/stepfun-ai/Step-3.5-Flash" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 31.6, metric: "index", source: "https://openrouter.ai/stepfun/step-3.5-flash/benchmarks", date: "2026-06-02" }, { name: "SciCode", score: 40.4, metric: "percent correct", source: "https://openrouter.ai/stepfun/step-3.5-flash/benchmarks", date: "2026-06-02" }, { name: "Terminal-Bench Hard", score: 27.3, metric: "success rate", source: "https://openrouter.ai/stepfun/step-3.5-flash/benchmarks", date: "2026-06-02" }, { name: "SWE-Bench Verified", score: 74.4, metric: "resolved", source: "https://arxiv.org/abs/2602.10604" }] }, "stepfun/step-3.5-flash-2603": { id: "stepfun/step-3.5-flash-2603", name: "Step 3.5 Flash 2603", description: "StepFun flash model for efficient multimodal reasoning, coding, and tool use", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-01", release_date: "2026-04-02", last_updated: "2026-04-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, input: 256e3, output: 256e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/stepfun-ai/Step-3.5-Flash" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 34.6, metric: "index", source: "https://openrouter.ai/stepfun/step-3.5-flash/benchmarks", date: "2026-06-02" }, { name: "SciCode", score: 38.5, metric: "percent correct", source: "https://openrouter.ai/stepfun/step-3.5-flash/benchmarks", date: "2026-06-02" }, { name: "Terminal-Bench Hard", score: 32.6, metric: "success rate", source: "https://openrouter.ai/stepfun/step-3.5-flash/benchmarks", date: "2026-06-02" }] }, "xiaomi/mimo-v2.5-pro-ultraspeed": { id: "xiaomi/mimo-v2.5-pro-ultraspeed", name: "MiMo-V2.5-Pro-UltraSpeed", description: "MiMo pro model for strong multimodal reasoning and agent execution", family: "mimo", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-12", release_date: "2026-06-08", last_updated: "2026-06-09", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro-FP4-DFlash" }] }, "xiaomi/mimo-v2-flash": { id: "xiaomi/mimo-v2-flash", name: "MiMo-V2-Flash", description: "MiMo flash model for fast multimodal assistance and agent workflows", family: "mimo", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-12-01", release_date: "2025-12-16", last_updated: "2026-02-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash" }] }, "xiaomi/mimo-v2-pro": { id: "xiaomi/mimo-v2-pro", name: "MiMo-V2-Pro", description: "Earlier MiMo Pro model for multimodal agents, reasoning, and code tasks", family: "mimo", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-12", release_date: "2026-03-18", last_updated: "2026-03-18", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 131072 } }, "xiaomi/mimo-v2-omni": { id: "xiaomi/mimo-v2-omni", name: "MiMo-V2-Omni", description: "MiMo omni model for text, image, video, audio, and agents", family: "mimo", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-12", release_date: "2026-03-18", last_updated: "2026-03-18", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 131072 } }, "xiaomi/mimo-v2.5": { id: "xiaomi/mimo-v2.5", name: "MiMo-V2.5", description: "Open MiMo model for multimodal coding agents and long-context automation", family: "mimo", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-12", release_date: "2026-04-22", last_updated: "2026-04-22", modalities: { input: ["text", "image", "audio", "video"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/XiaomiMiMo/MiMo-V2.5" }] }, "xiaomi/mimo-v2.5-pro": { id: "xiaomi/mimo-v2.5-pro", name: "MiMo-V2.5-Pro", description: "Stronger MiMo Pro tier for multimodal reasoning and coding-agent execution", family: "mimo", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-12", release_date: "2026-04-22", last_updated: "2026-04-22", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro" }], benchmarks: [{ name: "SWE-Bench Verified", score: 78.9, metric: "resolved", source: "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro" }, { name: "SWE-Bench Pro", score: 57.2, metric: "resolve rate", source: "https://mimo.xiaomi.com/mimo-v2-5-pro/", date: "2026-04-22" }, { name: "GPQA Diamond", score: 86.6, metric: "accuracy", source: "https://mimo.xiaomi.com/mimo-v2-5-pro/", date: "2026-04-22" }] }, "ibm/granite-4-h-micro": { id: "ibm/granite-4-h-micro", name: "Granite-4.0-H-Micro", description: "Compact open-weight hybrid Granite model for lightweight enterprise chat and tool calling", family: "granite", attachment: false, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2025-10-02", last_updated: "2025-10-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/ibm-granite/granite-4.0-h-micro" }] }, "ibm/granite-4-h-small": { id: "ibm/granite-4-h-small", name: "Granite-4.0-H-Small", description: "Open-weight hybrid model for enterprise chat, coding, retrieval-augmented generation, and tool-calling workloads", family: "granite", attachment: false, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2025-10-02", last_updated: "2025-10-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/ibm-granite/granite-4.0-h-small" }] }, "minimax/MiniMax-M2": { id: "minimax/MiniMax-M2", name: "MiniMax-M2", description: "Efficient open MiniMax model built for coding agents and tool-heavy workflows", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-10-27", last_updated: "2025-10-27", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M2" }], benchmarks: [{ name: "SWE-Bench Verified", score: 69.4, metric: "resolved", source: "https://huggingface.co/MiniMaxAI/MiniMax-M2" }] }, "minimax/MiniMax-M2-Her": { id: "minimax/MiniMax-M2-Her", name: "MiniMax-M2 Her", description: "MiniMax M2 variant tuned for conversational and character-driven agent interactions", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-01-23", last_updated: "2026-01-23", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 65536, output: 2048 } }, "minimax/MiniMax-M2.1": { id: "minimax/MiniMax-M2.1", name: "MiniMax-M2.1", description: "Earlier MiniMax agent model for practical coding and productivity tasks", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-12-23", last_updated: "2025-12-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M2.1" }], benchmarks: [{ name: "SWE-Bench Verified", score: 74, metric: "resolved", source: "https://huggingface.co/MiniMaxAI/MiniMax-M2.1" }, { name: "SWE-Bench Pro", score: 36.81, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "minimax/MiniMax-M2.5": { id: "minimax/MiniMax-M2.5", name: "MiniMax-M2.5", description: "Prior MiniMax coding model for agent workflows, office edits, and automation", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-02-12", last_updated: "2026-02-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M2.5" }], benchmarks: [{ name: "SWE-Bench Verified", score: 75.8, metric: "resolved", source: "https://www.swebench.com/" }, { name: "SWE-Atlas Codebase QnA", score: 10.3, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 19.52, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 18.6, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }] }, "minimax/MiniMax-M2.5-highspeed": { id: "minimax/MiniMax-M2.5-highspeed", name: "MiniMax-M2.5-highspeed", description: "High-speed MiniMax model for low-latency coding and agent workflows", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-02-13", last_updated: "2026-02-13", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M2.5" }] }, "minimax/image-01": { id: "minimax/image-01", name: "MiniMax image-01", description: "MiniMax text-to-image generation model with reference-image support", family: "minimax", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2025-02-15", last_updated: "2026-08-25", modalities: { input: ["text", "image"], output: ["image"] }, open_weights: false, limit: { context: 0, output: 0 } }, "minimax/MiniMax-M3": { id: "minimax/MiniMax-M3", name: "MiniMax-M3", description: "MiniMax multimodal model for long-context coding, perception, and agent planning", family: "minimax", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-01", last_updated: "2026-06-01", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 512e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M3" }], benchmarks: [{ name: "SWE-Bench Verified", score: 80.5, metric: "resolved", harness: "Claude Code", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "SWE-Bench Pro", score: 59, metric: "resolve rate", harness: "Claude Code", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "Terminal-Bench", score: 66, metric: "success rate", version: "2.1", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "BrowseComp", score: 83.52, metric: "accuracy", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "MCP Atlas", score: 74.2, metric: "success rate", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "OSWorld-Verified", score: 70.06, metric: "success rate", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }] }, "minimax/MiniMax-M2.7-highspeed": { id: "minimax/MiniMax-M2.7-highspeed", name: "MiniMax-M2.7-highspeed", description: "Low-latency M2.7 variant for interactive coding plans and agent loops", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-03-18", last_updated: "2026-03-18", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M2.7" }] }, "minimax/MiniMax-M2.7": { id: "minimax/MiniMax-M2.7", name: "MiniMax-M2.7", description: "Open MiniMax flagship for coding agents, office automation, and complex environments", family: "minimax", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-03-18", last_updated: "2026-03-18", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/MiniMaxAI/MiniMax-M2.7" }], benchmarks: [{ name: "SWE-Bench Verified", score: 79.9, metric: "resolved", harness: "Claude Code", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "SWE-Bench Pro", score: 56.2, metric: "resolve rate", harness: "Claude Code", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }, { name: "Terminal-Bench", score: 51.1, metric: "success rate", version: "2.1", source: "https://www.minimax.io/blog/minimax-m3", date: "2026-06-01" }] }, "alibaba/qwen3.7-max": { id: "alibaba/qwen3.7-max", name: "Qwen3.7 Max", description: "Qwen frontier model tuned for agent frameworks, coding assistants, and long tasks", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-05-21", last_updated: "2026-05-21", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 }, benchmarks: [{ name: "SWE-Bench Verified", score: 80.4, metric: "resolved", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "SWE-Bench Pro", score: 60.6, metric: "resolve rate", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "SWE-Bench Multilingual", score: 78.3, metric: "resolve rate", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "Terminal-Bench", score: 69.7, metric: "success rate", harness: "Terminus-2", version: "2.0", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "GPQA Diamond", score: 92.4, metric: "accuracy", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "Humanity's Last Exam", score: 41.4, metric: "accuracy", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "SciCode", score: 53.5, source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "MCP Atlas", score: 76.4, metric: "success rate", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }, { name: "NL2Repo", score: 47.2, harness: "Claude Code", source: "https://qwen.ai/blog?id=qwen3.7", date: "2026-05-19" }] }, "alibaba/qwen3-coder-plus": { id: "alibaba/qwen3-coder-plus", name: "Qwen3 Coder Plus", description: "Hosted Qwen coder for software agents, repo edits, and long-context code", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-07-23", last_updated: "2025-07-23", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "alibaba/qwen3-next-80b-a3b-thinking": { id: "alibaba/qwen3-next-80b-a3b-thinking", name: "Qwen3-Next 80B-A3B (Thinking)", description: "Efficient Qwen thinking model for local reasoning, math, and coding agents", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-09", last_updated: "2025-09", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking" }] }, "alibaba/qwen3.8-flash-next": { id: "alibaba/qwen3.8-flash-next", name: "Qwen3.8 Flash Next", description: "Open-weight experimental preview of the Qwen4 architecture: hybrid-attention MoE (125B total, 6B active) with vision encoder for coding, agent tasks, and image and video understanding", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-27", last_updated: "2026-08-27", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 131072 }, license: "qwen-community-1.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.8-Flash-Next" }] }, "alibaba/qwen3.5-9b": { id: "alibaba/qwen3.5-9b", name: "Qwen3.5 9B", description: "Qwen instruction model for multilingual chat, reasoning, and tool use", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-23", last_updated: "2026-02-23", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.5-9B" }] }, "alibaba/qwen-vl-max": { id: "alibaba/qwen-vl-max", name: "Qwen-VL Max", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2024-04-08", last_updated: "2025-08-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 8192 } }, "alibaba/qwen3-next-80b-a3b-instruct": { id: "alibaba/qwen3-next-80b-a3b-instruct", name: "Qwen3-Next 80B-A3B Instruct", description: "Qwen instruction model for multilingual chat, reasoning, and tool use", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-09", last_updated: "2025-09", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct" }] }, "alibaba/qwen3-coder-flash": { id: "alibaba/qwen3-coder-flash", name: "Qwen3 Coder Flash", description: "Qwen coding model for software agents, repository edits, and code reasoning", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-07-28", last_updated: "2025-07-28", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "alibaba/qwen3.8-max-preview": { id: "alibaba/qwen3.8-max-preview", name: "Qwen3.8 Max Preview", description: "Preview Qwen flagship for million-token multimodal reasoning and long-horizon agentic workflows", family: "qwen", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-07-19", last_updated: "2026-07-19", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 131072 }, benchmarks: [{ name: "Terminal-Bench", score: 86.6, metric: "accuracy", variant: "xhigh", version: "2.1", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "SWE-Bench Pro", score: 67.7, metric: "resolve rate", harness: "Claude Code", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "DeepSWE", score: 56.6, metric: "resolve rate", harness: "Claude Code", variant: "xhigh", version: "1.1", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "NL2Repo", score: 55.9, metric: "resolve rate", harness: "Claude Code", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "FrontierSWE", score: 73.5, metric: "dominance score", harness: "Claude Code", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "MLS-Bench-Lite", score: 41, metric: "score", harness: "Claude Code", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "AutomationBench", score: 27.3, metric: "pass@1", variant: "xhigh", dataset: "600-task public subset", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "Toolathlon Verified", score: 72.5, metric: "pass@1", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "WideSearch", score: 81.9, metric: "F1", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "Humanity's Last Exam", score: 56.2, metric: "accuracy", variant: "xhigh, with tools", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "GPQA Diamond", score: 92.6, metric: "accuracy", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "Humanity's Last Exam", score: 43.6, metric: "accuracy", variant: "xhigh, no tools", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "IFBench", score: 82.8, metric: "score", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "OSWorld-Verified", score: 86.1, metric: "success rate", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }, { name: "MMMU Pro", score: 82.3, metric: "accuracy", variant: "xhigh", source: "https://www.alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421", date: "2026-08-03" }] }, "alibaba/qwen-max": { id: "alibaba/qwen-max", name: "Qwen Max", description: "Flagship Qwen model for complex reasoning, coding, and agentic workflows", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2024-04-03", last_updated: "2025-01-25", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 32768, output: 8192 }, benchmarks: [{ name: "Aider Polyglot", score: 21.8, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-01-28" }] }, "alibaba/qwen3.6-plus": { id: "alibaba/qwen3.6-plus", name: "Qwen3.6 Plus", description: "Earlier Qwen multimodal workhorse for million-token agent and document tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2026-04-02", last_updated: "2026-04-02", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "alibaba/qwen2.5-coder-0.5b": { id: "alibaba/qwen2.5-coder-0.5b", name: "Qwen2.5-Coder-0.5B", description: "Tiny open Qwen code model for lightweight completion and on-device coding", family: "qwen", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2024-11-12", last_updated: "2024-11-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 32768, output: 8192 }, license: "Apache 2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B" }] }, "alibaba/qwen-vl-plus": { id: "alibaba/qwen-vl-plus", name: "Qwen-VL Plus", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2024-01-25", last_updated: "2025-08-15", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 8192 } }, "alibaba/qwen3.5-27b": { id: "alibaba/qwen3.5-27b", name: "Qwen3.5 27B", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-23", last_updated: "2026-02-23", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.5-27B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 72.4, metric: "resolved", source: "https://huggingface.co/Qwen/Qwen3.5-27B" }] }, "alibaba/qwen3.8-27b": { id: "alibaba/qwen3.8-27b", name: "Qwen3.8 27B", description: "Dense 27B vision-language model for coding, agent tasks, and image and video understanding", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-14", last_updated: "2026-08-14", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.8-27B" }], benchmarks: [{ name: "SWE-bench Pro", score: 61.7, metric: "resolved", source: "https://huggingface.co/Qwen/Qwen3.8-27B" }] }, "alibaba/qwen3.5-35b-a3b": { id: "alibaba/qwen3.5-35b-a3b", name: "Qwen3.5 35B-A3B", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-23", last_updated: "2026-02-23", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.5-35B-A3B" }] }, "alibaba/qwen-flash": { id: "alibaba/qwen-flash", name: "Qwen Flash", description: "Efficient Qwen model for fast chat, extraction, and high-volume workloads", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2025-07-28", last_updated: "2025-07-28", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 32768 } }, "alibaba/qwen-turbo": { id: "alibaba/qwen-turbo", name: "Qwen Turbo", description: "Efficient Qwen model for fast chat, extraction, and high-volume workloads", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2024-11-01", last_updated: "2025-04-28", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 16384 } }, "alibaba/qwen3-32b": { id: "alibaba/qwen3-32b", name: "Qwen3 32B", description: "Dense open Qwen model for self-hosted chat, reasoning, and coding", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-04", last_updated: "2025-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-32B" }], benchmarks: [{ name: "Aider Polyglot", score: 40, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-08" }] }, "alibaba/qwq-plus": { id: "alibaba/qwq-plus", name: "QwQ Plus", description: "Qwen reasoning model for deliberate problem solving, math, and coding", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2025-03-05", last_updated: "2025-03-05", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 8192 } }, "alibaba/qwen3.5-flash": { id: "alibaba/qwen3.5-flash", name: "Qwen3.5 Flash", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-23", last_updated: "2026-02-23", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "alibaba/qwen3-vl-plus": { id: "alibaba/qwen3-vl-plus", name: "Qwen3-VL Plus", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-09-23", last_updated: "2025-09-23", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 32768 } }, "alibaba/qwen3-coder-next": { id: "alibaba/qwen3-coder-next", name: "Qwen3 Coder Next", description: "Open-weight Qwen coding model for agents, repository edits, and multi-turn tool use", family: "qwen", attachment: false, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-09", release_date: "2026-02-03", last_updated: "2026-02-03", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-Coder-Next" }] }, "alibaba/qwen3-coder-30b-a3b-instruct": { id: "alibaba/qwen3-coder-30b-a3b-instruct", name: "Qwen3-Coder 30B-A3B Instruct", description: "Smaller Qwen coder for efficient local agents and repo-level fixes", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-04", last_updated: "2025-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 19.4, metric: "index", source: "https://openrouter.ai/qwen/qwen3-coder-30b-a3b-instruct/benchmarks", date: "2026-06-02" }, { name: "SciCode", score: 27.8, metric: "percent correct", source: "https://openrouter.ai/qwen/qwen3-coder-30b-a3b-instruct/benchmarks", date: "2026-06-02" }, { name: "Terminal-Bench Hard", score: 15.2, metric: "success rate", source: "https://openrouter.ai/qwen/qwen3-coder-30b-a3b-instruct/benchmarks", date: "2026-06-02" }] }, "alibaba/qwen3.5-397b-a17b": { id: "alibaba/qwen3.5-397b-a17b", name: "Qwen3.5 397B-A17B", description: "Large open Qwen multimodal MoE for visual agents and long technical tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-15", last_updated: "2026-02-15", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.5-397B-A17B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 76.4, metric: "resolved", source: "https://huggingface.co/Qwen/Qwen3.5-397B-A17B" }] }, "alibaba/qwq-32b": { id: "alibaba/qwq-32b", name: "QwQ 32B", description: "Open reasoning model from the Qwen team for math, coding, and step-by-step problem solving", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2025-03-05", last_updated: "2025-03-05", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/QwQ-32B" }] }, "alibaba/qwen3.6-27b": { id: "alibaba/qwen3.6-27b", name: "Qwen3.6 27B", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-22", last_updated: "2026-04-22", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.6-27B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 77.2, metric: "resolved", source: "https://huggingface.co/Qwen/Qwen3.6-27B" }] }, "alibaba/qwen3.7-flash": { id: "alibaba/qwen3.7-flash", name: "Qwen3.7 Flash", description: "Lightweight multimodal Qwen model for high-throughput text, image, and video tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-07-15", last_updated: "2026-07-15", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, input: 991e3, output: 65536 } }, "alibaba/qwen2-5-vl-72b-instruct": { id: "alibaba/qwen2-5-vl-72b-instruct", name: "Qwen2.5-VL 72B Instruct", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2024-09", last_updated: "2024-09", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct" }] }, "alibaba/qwen3.6-35b-a3b": { id: "alibaba/qwen3.6-35b-a3b", name: "Qwen3.6 35B-A3B", description: "Open multimodal Qwen MoE for local agents that need vision, audio, and code", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-17", last_updated: "2026-04-17", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.6-35B-A3B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 73.4, metric: "resolved", source: "https://huggingface.co/Qwen/Qwen3.6-35B-A3B" }] }, "alibaba/qwen3.8-max-0902": { id: "alibaba/qwen3.8-max-0902", name: "Qwen3.8 Max 0902", description: "2026-09-02 upgraded snapshot of Qwen3.8 Max with stronger coding, collaborative agents, and multimodal document understanding", family: "qwen", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-09-02", last_updated: "2026-09-02", modalities: { input: ["text", "image", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 131072 } }, "alibaba/qwen3-max": { id: "alibaba/qwen3-max", name: "Qwen3 Max", description: "Flagship Qwen3 model for coding agents, complex reasoning, and tool use", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-09-23", last_updated: "2025-09-23", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 65536 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 26.4, metric: "index", source: "https://openrouter.ai/qwen/qwen3-max/benchmarks", date: "2026-05-30" }, { name: "SciCode", score: 38.3, metric: "percent correct", source: "https://openrouter.ai/qwen/qwen3-max/benchmarks", date: "2026-05-30" }, { name: "Terminal-Bench Hard", score: 20.5, metric: "success rate", source: "https://openrouter.ai/qwen/qwen3-max/benchmarks", date: "2026-05-30" }] }, "alibaba/qwen3.8-2.4t-a95b": { id: "alibaba/qwen3.8-2.4t-a95b", name: "Qwen3.8 2.4T A95B", description: "Open-weight sparse MoE (2.4T total, 95B active), the open-weight twin of Qwen3.8 Max for coding, research, complex reasoning, and agentic workflows", family: "qwen", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-12", last_updated: "2026-08-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 131072 }, license: "qwen3.8-max", weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B" }] }, "alibaba/qwen-plus": { id: "alibaba/qwen-plus", name: "Qwen Plus", description: "Qwen instruction model for multilingual chat, reasoning, and tool use", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2024-01-25", last_updated: "2025-09-11", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 32768 } }, "alibaba/qwen3.5-122b-a10b": { id: "alibaba/qwen3.5-122b-a10b", name: "Qwen3.5 122B-A10B", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-23", last_updated: "2026-02-23", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3.5-122B-A10B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 72, metric: "resolved", source: "https://huggingface.co/Qwen/Qwen3.5-122B-A10B" }] }, "alibaba/qwen3.6-flash": { id: "alibaba/qwen3.6-flash", name: "Qwen3.6 Flash", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen3.6", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-27", last_updated: "2026-04-27", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "alibaba/qwen3-30b-a3b": { id: "alibaba/qwen3-30b-a3b", name: "Qwen3 30B A3B", description: "Sparse MoE Qwen model with 3B active parameters for efficient chat and reasoning", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-04-28", last_updated: "2025-04-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-30B-A3B" }] }, "alibaba/qwen3.8-flash": { id: "alibaba/qwen3.8-flash", name: "Qwen3.8 Flash", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, release_date: "2026-08-26", last_updated: "2026-08-26", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 131072 } }, "alibaba/qwen3.6-max-preview": { id: "alibaba/qwen3.6-max-preview", name: "Qwen3.6 Max Preview", description: "Flagship Qwen model for complex reasoning, coding, and agentic workflows", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2026-04-20", last_updated: "2026-04-20", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 65536 } }, "alibaba/qwen3.8-max": { id: "alibaba/qwen3.8-max", name: "Qwen3.8 Max", description: "2.4-trillion-parameter MoE flagship for coding, professional work, multimodal understanding, and long-horizon agentic workflows", family: "qwen", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-08-03", last_updated: "2026-08-03", modalities: { input: ["text", "image", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 131072 } }, "alibaba/qwen3-235b-a22b": { id: "alibaba/qwen3-235b-a22b", name: "Qwen3 235B-A22B", description: "Large open Qwen MoE for multilingual reasoning, coding, and tool use", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-04", last_updated: "2025-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-235B-A22B" }], benchmarks: [{ name: "Aider Polyglot", score: 59.6, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-09" }, { name: "SWE-Bench Pro", score: 21.41, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "alibaba/qwen3-vl-235b-a22b-thinking": { id: "alibaba/qwen3-vl-235b-a22b-thinking", name: "Qwen3 VL 235B A22B Thinking", description: "Qwen vision-language thinking model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-09-23", last_updated: "2025-09-23", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Thinking" }] }, "alibaba/qwen3-vl-235b-a22b-instruct": { id: "alibaba/qwen3-vl-235b-a22b-instruct", name: "Qwen3 VL 235B A22B Instruct", description: "Qwen vision-language instruct model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-09-23", last_updated: "2025-09-23", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Instruct" }] }, "alibaba/qwen3.7-plus": { id: "alibaba/qwen3.7-plus", name: "Qwen3.7 Plus", description: "Multimodal Qwen workhorse for long-context agents, visual inputs, and coding", family: "qwen", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2026-06-02", last_updated: "2026-06-02", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 64e3 } }, "alibaba/qwen3-235b-a22b-instruct-2507": { id: "alibaba/qwen3-235b-a22b-instruct-2507", name: "Qwen3 235B-A22B Instruct 2507", description: "Updated large open Qwen3 MoE instruct model for multilingual chat, coding, and tool use", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2025-07-21", last_updated: "2025-07-21", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 16384 }, license: "Apache 2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507" }] }, "alibaba/qwen-omni-turbo": { id: "alibaba/qwen-omni-turbo", name: "Qwen-Omni Turbo", description: "Qwen omni model for text, vision, audio, and multimodal agent tasks", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-04", release_date: "2025-01-19", last_updated: "2025-03-26", modalities: { input: ["text", "image", "audio", "video"], output: ["text", "audio"] }, open_weights: false, limit: { context: 32768, output: 2048 } }, "alibaba/qwen3-coder-480b-a35b-instruct": { id: "alibaba/qwen3-coder-480b-a35b-instruct", name: "Qwen3-Coder 480B-A35B Instruct", description: "Open Qwen coding heavyweight for repository reasoning and agentic engineering", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-04", last_updated: "2025-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 65536 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct" }], benchmarks: [{ name: "SWE-Bench Pro", score: 38.7, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "alibaba/qwen2.5-coder-32b-instruct": { id: "alibaba/qwen2.5-coder-32b-instruct", name: "Qwen2.5-Coder-32B-Instruct", description: "Open coding-focused Qwen model for code generation, repair, and repository reasoning", family: "qwen", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2024-11-12", last_updated: "2024-11-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct" }] }, "alibaba/qwen3.5-plus": { id: "alibaba/qwen3.5-plus", name: "Qwen3.5 Plus", description: "Qwen vision-language model for visual reasoning, documents, and agent tasks", family: "qwen", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2026-02-16", last_updated: "2026-02-16", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "nvidia/nemotron-3.5-content-safety": { id: "nvidia/nemotron-3.5-content-safety", name: "Nemotron 3.5 Content Safety", description: "Safety model for policy screening, moderation, and risk-aware routing workflows", family: "nemotron", attachment: true, reasoning: true, tool_call: false, temperature: true, release_date: "2026-06-04", last_updated: "2026-06-04", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "nvidia/mistral-nemotron": { id: "nvidia/mistral-nemotron", name: "Mistral Nemotron", description: "Mistral model for multilingual chat, reasoning, and tool-assisted workflows", family: "nemotron", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2025-06-11", last_updated: "2025-06-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "nvidia/llama-3.1-nemotron-ultra-253b": { id: "nvidia/llama-3.1-nemotron-ultra-253b", name: "Llama 3.1 Nemotron Ultra 253B", description: "Flagship Nemotron model for high-throughput reasoning and complex agents", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-04-07", last_updated: "2025-04-07", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "nvidia/nemotron-3.5-lightning": { id: "nvidia/nemotron-3.5-lightning", name: "Nemotron 3.5 Lightning 30B A3B", description: "Fast NVIDIA Nemotron MoE for reliable agentic tasks across enterprise workloads", family: "nemotron", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-11", last_updated: "2026-08-11", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 } }, "nvidia/llama-3.3-nemotron-super-49b-v1.5": { id: "nvidia/llama-3.3-nemotron-super-49b-v1.5", name: "Llama 3.3 Nemotron Super 49B v1.5", description: "Nemotron model for efficient reasoning, coding, and specialized AI agents", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-07-25", last_updated: "2025-07-25", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 } }, "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning": { id: "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning", name: "Nemotron 3 Nano Omni 30B A3B Reasoning", description: "Open Nemotron omni model combining reasoning with text, vision, and audio", family: "nemotron", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-04-28", last_updated: "2026-04-28", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 65536 } }, "nvidia/nemotron-voicechat": { id: "nvidia/nemotron-voicechat", name: "Nemotron VoiceChat", description: "Nemotron multimodal model for visual reasoning and agentic AI workflows", family: "nemotron", attachment: true, reasoning: false, tool_call: true, temperature: true, release_date: "2026-03-16", last_updated: "2026-03-16", modalities: { input: ["text", "audio"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "nvidia/nemotron-nano-9b-v2": { id: "nvidia/nemotron-nano-9b-v2", name: "Nemotron Nano 9B v2", description: "Compact Nemotron model for efficient reasoning and deployable AI agents", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-08-18", last_updated: "2025-08-18", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 } }, "nvidia/nemotron-3-content-safety": { id: "nvidia/nemotron-3-content-safety", name: "Nemotron 3 Content Safety", description: "Safety model for policy screening, moderation, and risk-aware routing workflows", family: "nemotron", attachment: false, reasoning: false, tool_call: false, temperature: false, release_date: "2026-04-16", last_updated: "2026-04-16", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 } }, "nvidia/nemotron-nano-12b-v2-vl": { id: "nvidia/nemotron-nano-12b-v2-vl", name: "Nemotron Nano 12B v2 VL", description: "Nemotron multimodal model for visual reasoning and agentic AI workflows", family: "nemotron", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2025-10-28", last_updated: "2025-10-28", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 } }, "nvidia/llama-nemotron-embed-vl-1b-v2": { id: "nvidia/llama-nemotron-embed-vl-1b-v2", name: "Llama Nemotron Embed VL 1B v2", description: "Embedding model for semantic search, retrieval, clustering, and ranking pipelines", family: "nemotron", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2026-02-10", last_updated: "2026-02-10", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 32768, output: 2048 } }, "nvidia/nemotron-3-super-120b-a12b": { id: "nvidia/nemotron-3-super-120b-a12b", name: "Nemotron 3 Super 120B A12B", description: "Nemotron middle tier for collaborative agents and high-volume reasoning workloads", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-03-11", last_updated: "2026-03-11", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 } }, "nvidia/llama-nemotron-rerank-vl-1b-v2": { id: "nvidia/llama-nemotron-rerank-vl-1b-v2", name: "Llama Nemotron Rerank VL 1B v2", description: "Reranking model for improving retrieval quality in search and recommendation systems", family: "nemotron", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2026-03-31", last_updated: "2026-03-31", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 } }, "nvidia/llama-3.1-nemotron-safety-guard-8b-v3": { id: "nvidia/llama-3.1-nemotron-safety-guard-8b-v3", name: "Llama 3.1 Nemotron Safety Guard 8B v3", description: "Safety model for policy screening, moderation, and risk-aware routing workflows", family: "nemotron", attachment: false, reasoning: false, tool_call: false, temperature: false, release_date: "2025-10-28", last_updated: "2025-10-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 } }, "nvidia/nemotron-mini-4b-instruct": { id: "nvidia/nemotron-mini-4b-instruct", name: "Nemotron Mini 4B Instruct", description: "Compact Nemotron model for efficient reasoning and deployable AI agents", family: "nemotron", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2024-08-21", last_updated: "2024-08-26", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "nvidia/nemotron-content-safety-reasoning-4b": { id: "nvidia/nemotron-content-safety-reasoning-4b", name: "Nemotron Content Safety Reasoning 4B", description: "Safety model for policy screening, moderation, and risk-aware routing workflows", family: "nemotron", attachment: false, reasoning: true, tool_call: false, temperature: false, release_date: "2026-01-22", last_updated: "2026-01-22", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 } }, "nvidia/nemotron-cascade-2-30b-a3b": { id: "nvidia/nemotron-cascade-2-30b-a3b", name: "Nemotron Cascade 2 30B A3B", description: "Nemotron model for efficient reasoning, coding, and specialized AI agents", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-03-24", last_updated: "2026-04-09", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 32768 } }, "nvidia/nemotron-3-ultra-550b-a55b": { id: "nvidia/nemotron-3-ultra-550b-a55b", name: "Nemotron 3 Ultra 550B A55B", description: "Largest Nemotron 3 model for maximum open-weight reasoning and agent accuracy", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-06-04", last_updated: "2026-06-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Verified", score: 70.7, metric: "resolved", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "SWE-Bench Multilingual", score: 67.7, metric: "resolve rate", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "Terminal-Bench", score: 56.4, metric: "success rate", version: "2.1", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "GPQA", score: 87, metric: "accuracy", variant: "no tools", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "Humanity's Last Exam", score: 26.7, metric: "accuracy", variant: "no tools", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "Humanity's Last Exam", score: 37.4, metric: "accuracy", variant: "with tools", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "LiveCodeBench", score: 89, metric: "pass@1", version: "v6", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "MMLU-Pro", score: 86.8, metric: "accuracy", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "BrowseComp", score: 44.4, metric: "accuracy", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "IFBench", score: 81.7, metric: "accuracy", variant: "prompt loose", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }, { name: "GDPval", score: 46.7, metric: "wins or ties", source: "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", date: "2026-06-04" }] }, "nvidia/llama-3.3-nemotron-super-49b-v1": { id: "nvidia/llama-3.3-nemotron-super-49b-v1", name: "Llama 3.3 Nemotron Super 49B v1", description: "Nemotron model for efficient reasoning, coding, and specialized AI agents", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-04-07", last_updated: "2025-04-07", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 } }, "nvidia/nemotron-3-nano-30b-a3b": { id: "nvidia/nemotron-3-nano-30b-a3b", name: "Nemotron 3 Nano 30B A3B", description: "Small Nemotron 3 MoE for efficient coding, math, and long-context agents", family: "nemotron", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-12-15", last_updated: "2025-12-15", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 } }, "nvidia/llama-3.1-nemotron-70b-instruct": { id: "nvidia/llama-3.1-nemotron-70b-instruct", name: "Llama 3.1 Nemotron 70B Instruct", description: "Nemotron model for efficient reasoning, coding, and specialized AI agents", family: "nemotron", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2025-04-15", last_updated: "2025-04-15", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "anthropic/claude-sonnet-4-6": { id: "anthropic/claude-sonnet-4-6", name: "Claude Sonnet 4.6", description: "Claude workhorse for coding agents, careful analysis, and production cost control", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-08-31", release_date: "2026-02-17", last_updated: "2026-03-13", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 64e3 }, benchmarks: [{ name: "SWE-Atlas Codebase QnA", score: 31.2, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 32.21, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 31.76, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "Artificial Analysis Coding Agent Index", score: 49.4, metric: "average pass@1", harness: "Claude Code", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 70.3, metric: "pass@1", harness: "Claude Code", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 14.9, metric: "pass@1", harness: "Claude Code", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 63.1, metric: "pass@1", harness: "Claude Code", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 67, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "Humanity's Last Exam", score: 34.6, metric: "accuracy", variant: "no tools", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "Humanity's Last Exam", score: 46.8, metric: "accuracy", variant: "with tools", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "OSWorld-Verified", score: 78.5, metric: "success rate", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }] }, "anthropic/claude-3-haiku-20240307": { id: "anthropic/claude-3-haiku-20240307", name: "Claude Haiku 3", description: "Legacy model retained for compatibility with older integrations", family: "claude-haiku", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2023-08-31", release_date: "2024-03-13", last_updated: "2024-03-13", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 4096 } }, "anthropic/claude-opus-5": { id: "anthropic/claude-opus-5", name: "Claude Opus 5", description: "Strongest Claude Opus model for coding, agents, and professional work", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2026-05", release_date: "2026-07-24", last_updated: "2026-07-24", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Verified", score: 96, metric: "resolved", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "SWE-Bench Pro", score: 79.2, metric: "resolve rate", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "SWE-Bench Multilingual", score: 89.5, metric: "resolve rate", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "SWE-Bench Multimodal", score: 59.4, metric: "resolve rate", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "DeepSWE", score: 68.8, metric: "resolve rate", version: "1.1", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "FrontierCode", score: 53.4, metric: "mean@5", variant: "medium effort", dataset: "Main", version: "1.1", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "Frontier-Bench", score: 43.3, metric: "mean reward", harness: "mini-SWE-agent", variant: "max effort", version: "v0.1", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "BrowseComp", score: 90.8, metric: "accuracy", variant: "single agent", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "Humanity's Last Exam", score: 56.3, metric: "accuracy", variant: "no tools", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "Humanity's Last Exam", score: 64.7, metric: "accuracy", variant: "with tools", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "DeepSearchQA", score: 95, metric: "F1", variant: "max effort", source: "https://www.anthropic.com/news/claude-opus-5", date: "2026-07-24" }, { name: "OSWorld", score: 70.6, metric: "success rate", version: "2.0", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "GDPval-AA", score: 1861, metric: "Elo", variant: "max effort", version: "v2", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "AA-Briefcase", score: 1720, metric: "Elo", variant: "max effort", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "AutomationBench", score: 26, metric: "success rate", variant: "max effort", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "ARC-AGI-1", score: 97.5, metric: "accuracy", variant: "max effort", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "ARC-AGI-2", score: 90.4, metric: "accuracy", variant: "max effort", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "ARC-AGI-3", score: 30.2, metric: "RHAE", variant: "high effort", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }, { name: "HealthBench Professional", score: 59.8, metric: "score", variant: "max effort", source: "https://www.anthropic.com/claude-opus-5-system-card", date: "2026-07-24" }] }, "anthropic/claude-3-7-sonnet-20250219": { id: "anthropic/claude-3-7-sonnet-20250219", name: "Claude Sonnet 3.7", description: "Balanced Claude model for coding, analysis, agent workflows, and cost control", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-10-31", release_date: "2025-02-19", last_updated: "2025-02-19", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 }, benchmarks: [{ name: "Aider Polyglot", score: 64.9, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-02-24" }] }, "anthropic/claude-opus-4-5": { id: "anthropic/claude-opus-4-5", name: "Claude Opus 4.5 (latest)", description: "Flagship Claude model for deep reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-05", release_date: "2025-11-24", last_updated: "2025-11-24", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 } }, "anthropic/claude-sonnet-4-0": { id: "anthropic/claude-sonnet-4-0", name: "Claude Sonnet 4 (latest)", description: "Balanced Claude model for coding, analysis, agent workflows, and cost control", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-05-22", last_updated: "2025-05-22", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 }, benchmarks: [{ name: "Aider Polyglot", score: 61.3, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-24" }, { name: "SWE-Bench Pro", score: 42.7, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "anthropic/claude-opus-4-1-20250805": { id: "anthropic/claude-opus-4-1-20250805", name: "Claude Opus 4.1", description: "Flagship Claude model for deep reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-08-05", last_updated: "2025-08-05", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 32e3 } }, "anthropic/claude-3-5-haiku-20241022": { id: "anthropic/claude-3-5-haiku-20241022", name: "Claude Haiku 3.5", description: "Fast Claude model for responsive assistance, classification, and lightweight agents", family: "claude-haiku", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-07-31", release_date: "2024-10-22", last_updated: "2024-10-22", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 8192 }, benchmarks: [{ name: "Aider Polyglot", score: 28, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2024-12-21" }] }, "anthropic/claude-fable-5-1": { id: "anthropic/claude-fable-5-1", name: "Claude Fable 5.1", description: "Claude model for demanding reasoning and long-horizon agentic work", family: "claude-fable", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2026-06", release_date: "2026-09-01", last_updated: "2026-09-01", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 } }, "anthropic/claude-opus-4-20250514": { id: "anthropic/claude-opus-4-20250514", name: "Claude Opus 4", description: "Flagship Claude model for deep reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-05-22", last_updated: "2025-05-22", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 32e3 }, benchmarks: [{ name: "Aider Polyglot", score: 72, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-25" }] }, "anthropic/claude-opus-4-0": { id: "anthropic/claude-opus-4-0", name: "Claude Opus 4 (latest)", description: "Flagship Claude model for deep reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-05-22", last_updated: "2025-05-22", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 32e3 }, benchmarks: [{ name: "Aider Polyglot", score: 72, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-25" }] }, "anthropic/claude-opus-4-6": { id: "anthropic/claude-opus-4-6", name: "Claude Opus 4.6", description: "High-end Claude for difficult coding, planning, and slower expert reasoning", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-05-31", release_date: "2026-02-05", last_updated: "2026-03-13", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 51.9, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }, { name: "SWE-Atlas Codebase QnA", score: 33.3, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Codebase QnA", score: 30, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 35.58, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 36.67, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "SWE-Atlas Test Writing", score: 36.08, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "Artificial Analysis Coding Agent Index", score: 51.3, metric: "average pass@1", harness: "Claude Code", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 71.9, metric: "pass@1", harness: "Claude Code", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 11.8, metric: "pass@1", harness: "Claude Code", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 70.2, metric: "pass@1", harness: "Claude Code", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }] }, "anthropic/claude-sonnet-4-5-20250929": { id: "anthropic/claude-sonnet-4-5-20250929", name: "Claude Sonnet 4.5", description: "Balanced Claude model for coding, analysis, agent workflows, and cost control", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-07-31", release_date: "2025-09-29", last_updated: "2025-09-29", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 } }, "anthropic/claude-opus-4-7": { id: "anthropic/claude-opus-4-7", name: "Claude Opus 4.7", description: "Stronger Opus tier for advanced software work and high-stakes reasoning", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2026-01-31", release_date: "2026-04-16", last_updated: "2026-04-16", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 64.3, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "Terminal-Bench", score: 66.1, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "SWE-Atlas Refactoring", score: 48.57, metric: "score", harness: "Claude Code", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "Artificial Analysis Coding Agent Index", score: 66.6, metric: "average pass@1", harness: "Claude Code", variant: "max", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 81, metric: "pass@1", harness: "Claude Code", variant: "max", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 44.9, metric: "pass@1", harness: "Claude Code", variant: "max", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 73.8, metric: "pass@1", harness: "Claude Code", variant: "max", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Artificial Analysis Coding Agent Index", score: 61.2, metric: "average pass@1", harness: "Cursor CLI", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 78.4, metric: "pass@1", harness: "Cursor CLI", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 34.4, metric: "pass@1", harness: "Cursor CLI", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 70.6, metric: "pass@1", harness: "Cursor CLI", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Artificial Analysis Coding Agent Index", score: 59.9, metric: "average pass@1", harness: "Claude Code", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 71.7, metric: "pass@1", harness: "Claude Code", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 36.4, metric: "pass@1", harness: "Claude Code", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 71.4, metric: "pass@1", harness: "Claude Code", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "GPQA Diamond", score: 94.2, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 46.9, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 54.7, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "OSWorld-Verified", score: 78, metric: "success rate", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }] }, "anthropic/claude-haiku-4-5-20251001": { id: "anthropic/claude-haiku-4-5-20251001", name: "Claude Haiku 4.5", description: "Fast Claude model for responsive assistance, classification, and lightweight agents", family: "claude-haiku", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-02-28", release_date: "2025-10-15", last_updated: "2025-10-15", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 } }, "anthropic/claude-fable-5": { id: "anthropic/claude-fable-5", name: "Claude Fable 5", description: "Claude model for creative writing, analysis, and controlled agent workflows", family: "claude-fable", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2026-01-31", release_date: "2026-06-09", last_updated: "2026-06-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 80.3, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "SWE-Bench Verified", score: 95, metric: "resolved", source: "https://benchlm.ai/benchmarks/sweVerified" }, { name: "Terminal-Bench", score: 88, metric: "success rate", version: "2.1", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "Humanity's Last Exam", score: 59, metric: "accuracy", variant: "no tools", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "Humanity's Last Exam", score: 64.5, metric: "accuracy", variant: "with tools", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "OSWorld-Verified", score: 85, metric: "success rate", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "FrontierCode", score: 29.3, metric: "pass rate", variant: "high effort", dataset: "Diamond", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "GDPval-AA", score: 1932, metric: "Elo", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "AutomationBench", score: 17.4, metric: "success rate", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }] }, "anthropic/claude-haiku-4-5": { id: "anthropic/claude-haiku-4-5", name: "Claude Haiku 4.5 (latest)", description: "Fast Claude lane for lightweight agents, office tasks, and responsive chat", family: "claude-haiku", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-02-28", release_date: "2025-10-15", last_updated: "2025-10-15", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 39.45, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "anthropic/claude-sonnet-4-5": { id: "anthropic/claude-sonnet-4-5", name: "Claude Sonnet 4.5 (latest)", description: "Balanced Claude model for coding, analysis, agent workflows, and cost control", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-07-31", release_date: "2025-09-29", last_updated: "2025-09-29", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 43.6, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "anthropic/claude-opus-4-1": { id: "anthropic/claude-opus-4-1", name: "Claude Opus 4.1 (latest)", description: "Flagship Claude model for deep reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-08-05", last_updated: "2025-08-05", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 32e3 } }, "anthropic/claude-opus-4-8": { id: "anthropic/claude-opus-4-8", name: "Claude Opus 4.8", description: "Top Claude Opus tier for the hardest reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2026-01", release_date: "2026-05-28", last_updated: "2026-05-28", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 69.2, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "Terminal-Bench", score: 74.6, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "SWE-Bench Verified", score: 88.6, metric: "resolved", source: "https://benchlm.ai/benchmarks/sweVerified" }, { name: "Humanity's Last Exam", score: 49.8, metric: "accuracy", variant: "no tools", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "Humanity's Last Exam", score: 57.9, metric: "accuracy", variant: "with tools", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "OSWorld-Verified", score: 83.4, metric: "success rate", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }, { name: "FrontierCode", score: 13.4, metric: "pass rate", variant: "high effort", dataset: "Diamond", source: "https://www.anthropic.com/news/claude-fable-5-mythos-5", date: "2026-06-09" }] }, "anthropic/claude-sonnet-4-20250514": { id: "anthropic/claude-sonnet-4-20250514", name: "Claude Sonnet 4", description: "Balanced Claude model for coding, analysis, agent workflows, and cost control", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03-31", release_date: "2025-05-22", last_updated: "2025-05-22", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 }, benchmarks: [{ name: "Aider Polyglot", score: 61.3, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-24" }] }, "anthropic/claude-3-5-sonnet-20241022": { id: "anthropic/claude-3-5-sonnet-20241022", name: "Claude Sonnet 3.5 v2", description: "Balanced Claude model for coding, analysis, agent workflows, and cost control", family: "claude-sonnet", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-04-30", release_date: "2024-10-22", last_updated: "2024-10-22", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 8192 }, benchmarks: [{ name: "Aider Polyglot", score: 51.6, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-01-17" }] }, "anthropic/claude-sonnet-5": { id: "anthropic/claude-sonnet-5", name: "Claude Sonnet 5", description: "Everyday Claude agent model for coding, planning, browsing, and general work", family: "claude-sonnet", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2026-01-31", release_date: "2026-06-30", last_updated: "2026-06-30", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Verified", score: 85.2, metric: "resolved", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "SWE-Bench Pro", score: 63.2, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "SWE-Bench Multilingual", score: 78.3, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "Terminal-Bench", score: 80.4, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "OSWorld-Verified", score: 81.2, metric: "success rate", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "BrowseComp", score: 84.7, metric: "accuracy", variant: "single agent", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }, { name: "FrontierCode", score: 38.8, metric: "pass rate", version: "v1", source: "https://www.anthropic.com/news/claude-sonnet-5", date: "2026-06-30" }] }, "anthropic/claude-mythos-5": { id: "anthropic/claude-mythos-5", name: "Claude Mythos 5", description: "Restricted Claude model for advanced cybersecurity and biology research workflows", family: "claude-mythos", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2026-01-31", release_date: "2026-06-09", last_updated: "2026-06-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 128e3 } }, "anthropic/claude-opus-4-5-20251101": { id: "anthropic/claude-opus-4-5-20251101", name: "Claude Opus 4.5", description: "Flagship Claude model for deep reasoning, coding, and long-horizon agents", family: "claude-opus", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-05", release_date: "2025-11-01", last_updated: "2025-11-01", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 64e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 45.89, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "google/gemma-4-26b-a4b-it": { id: "google/gemma-4-26b-a4b-it", name: "Gemma 4 26B A4B IT", description: "Open Gemma instruction model for efficient chat and self-hosted deployments", family: "gemma", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-02", last_updated: "2026-04-02", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/google/gemma-4-26B-A4B-it" }] }, "google/gemini-2.5-flash-tts": { id: "google/gemini-2.5-flash-tts", name: "Gemini 2.5 Flash TTS", description: "Speech generation model for controllable voice, narration, and audio delivery", family: "gemini-flash", attachment: false, reasoning: false, tool_call: false, temperature: true, knowledge: "2025-01", release_date: "2025-09-30", last_updated: "2025-12-10", modalities: { input: ["text"], output: ["audio"] }, open_weights: false, limit: { context: 32768, output: 16384 } }, "google/gemini-3.1-pro-preview-customtools": { id: "google/gemini-3.1-pro-preview-customtools", name: "Gemini 3.1 Pro Preview Custom Tools", description: "Advanced Gemini model for complex reasoning, coding, and multimodal analysis", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-02-19", last_updated: "2026-02-19", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemini-2.5-pro-tts": { id: "google/gemini-2.5-pro-tts", name: "Gemini 2.5 Pro TTS", description: "Speech generation model for controllable voice, narration, and audio delivery", family: "gemini-pro", attachment: false, reasoning: false, tool_call: false, temperature: false, knowledge: "2025-01", release_date: "2025-09-30", last_updated: "2025-12-10", modalities: { input: ["text"], output: ["audio"] }, open_weights: false, limit: { context: 32768, output: 16384 } }, "google/gemini-3.1-flash-lite-image": { id: "google/gemini-3.1-flash-lite-image", name: "Nano Banana 2 Lite", description: "Fastest, most cost-efficient Gemini image model for high-volume 1K generation and editing", family: "gemini-flash-lite", attachment: true, reasoning: true, tool_call: true, structured_output: false, temperature: true, knowledge: "2025-01", release_date: "2026-06-30", last_updated: "2026-06-30", modalities: { input: ["text", "image"], output: ["text", "image"] }, open_weights: false, limit: { context: 65536, output: 4096 } }, "google/gemma-4-E2B-it": { id: "google/gemma-4-E2B-it", name: "Gemma 4 E2B IT", description: "Open Gemma instruction model for efficient chat and self-hosted deployments", family: "gemma", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-02", last_updated: "2026-04-02", modalities: { input: ["text", "image", "audio"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/google/gemma-4-E2B-it" }] }, "google/gemma-3-4b-it": { id: "google/gemma-3-4b-it", name: "Gemma 3 4B IT", description: "Open multimodal Gemma instruction model for efficient text generation and image understanding", family: "gemma", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-08", release_date: "2025-03-12", last_updated: "2025-03-12", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/google/gemma-3-4b-it" }] }, "google/gemini-3.5-transcribe-live": { id: "google/gemini-3.5-transcribe-live", name: "Gemini 3.5 Transcribe Live", description: "Speech transcription model for accurate audio-to-text and captioning workflows", family: "gemini", attachment: false, reasoning: false, tool_call: false, release_date: "2026-08-26", last_updated: "2026-08-26", modalities: { input: ["audio"], output: ["text"] }, open_weights: false, limit: { context: 0, output: 0 } }, "google/lyria-3-clip-preview": { id: "google/lyria-3-clip-preview", name: "Lyria 3 Clip Preview", description: "Music generation model for short 30-second clips, loops, and previews from text or image prompts", family: "lyria", attachment: true, reasoning: false, tool_call: false, structured_output: false, temperature: true, release_date: "2026-03-25", last_updated: "2026-03-25", modalities: { input: ["text", "image"], output: ["text", "audio"] }, open_weights: false, limit: { context: 131072, output: 65536 } }, "google/gemini-2.5-flash-image": { id: "google/gemini-2.5-flash-image", name: "Nano Banana", description: "Nano Banana image model for fast generation, edits, and character-consistent assets", family: "gemini-flash", attachment: true, reasoning: true, tool_call: false, temperature: true, knowledge: "2024-06", release_date: "2025-08-26", last_updated: "2025-08-26", modalities: { input: ["text", "image"], output: ["text", "image"] }, open_weights: false, limit: { context: 32768, output: 32768 } }, "google/deep-research-max-preview-04-2026": { id: "google/deep-research-max-preview-04-2026", name: "Deep Research Max Preview", description: "Maximum-comprehensiveness agentic researcher for multi-step investigation, synthesis, and cited reports", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2025-01", release_date: "2026-04-21", last_updated: "2026-04-21", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text", "image"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemini-2.0-flash": { id: "google/gemini-2.0-flash", name: "Gemini 2.0 Flash", description: "Earlier Gemini Flash workhorse for responsive multimodal apps and tool use", family: "gemini-flash", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-06", release_date: "2024-12-11", last_updated: "2024-12-11", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 8192 } }, "google/gemini-3-pro-image": { id: "google/gemini-3-pro-image", name: "Nano Banana Pro", description: "Nano Banana Pro for higher-fidelity image generation and design-heavy edits", family: "gemini-pro", attachment: true, reasoning: true, tool_call: false, temperature: true, knowledge: "2025-01", release_date: "2026-05-28", last_updated: "2026-05-28", modalities: { input: ["text", "image"], output: ["text", "image"] }, open_weights: false, limit: { context: 65536, output: 32768 } }, "google/gemma-4-E4B-it": { id: "google/gemma-4-E4B-it", name: "Gemma 4 E4B IT", description: "Open Gemma instruction model for efficient chat and self-hosted deployments", family: "gemma", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-02", last_updated: "2026-04-02", modalities: { input: ["text", "image", "audio"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/google/gemma-4-E4B-it" }] }, "google/gemini-3.1-pro-preview": { id: "google/gemini-3.1-pro-preview", name: "Gemini 3.1 Pro Preview", description: "Reasoning-first Gemini preview for agentic coding and complex problem solving", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-02-19", last_updated: "2026-02-19", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "SWE-Bench Pro", score: 54.2, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "Terminal-Bench", score: 70.3, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "SWE-Bench Pro", score: 46.1, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }, { name: "SWE-Atlas Codebase QnA", score: 13.5, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 33.81, metric: "score", harness: "Gemini CLI", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 29.84, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "Artificial Analysis Coding Agent Index", score: 43, metric: "average pass@1", harness: "Gemini CLI", variant: "high", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 45.6, metric: "pass@1", harness: "Gemini CLI", variant: "high", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 15.1, metric: "pass@1", harness: "Gemini CLI", variant: "high", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 68.3, metric: "pass@1", harness: "Gemini CLI", variant: "high", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "GPQA Diamond", score: 94.3, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 44.4, metric: "accuracy", dataset: "full set, text + MM", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "ARC-AGI-2", score: 77.1, metric: "accuracy", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "MMMU Pro", score: 80.5, metric: "accuracy", variant: "no tools", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "MCP Atlas", score: 78.2, metric: "success rate", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "OSWorld-Verified", score: 76.2, metric: "success rate", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "CharXiv Reasoning", score: 83.3, metric: "accuracy", variant: "no tools", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "GDPval-AA", score: 1314, metric: "Elo", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }] }, "google/deep-research-preview-04-2026": { id: "google/deep-research-preview-04-2026", name: "Gemini Deep Research Preview", description: "Agentic model for autonomous multi-step research, synthesis, and cited reports", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2025-01", release_date: "2026-04-21", last_updated: "2026-04-21", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text", "image"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemini-2.5-flash-lite": { id: "google/gemini-2.5-flash-lite", name: "Gemini 2.5 Flash-Lite", description: "Lean Gemini 2.5 lane for cheap multimodal traffic and quick agents", family: "gemini-flash-lite", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2025-06-17", last_updated: "2025-06-17", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 9.5, metric: "index", source: "https://openrouter.ai/google/gemini-2.5-flash-lite/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 19.3, metric: "percent correct", source: "https://openrouter.ai/google/gemini-2.5-flash-lite/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 4.5, metric: "success rate", source: "https://openrouter.ai/google/gemini-2.5-flash-lite/benchmarks", date: "2026-03-11" }] }, "google/gemini-2.5-computer-use-preview-10-2025": { id: "google/gemini-2.5-computer-use-preview-10-2025", name: "Gemini 2.5 Computer Use Preview", description: "Specialized Gemini 2.5 model for browser-control agents that automate UI tasks", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-01", release_date: "2025-10-07", last_updated: "2025-10-07", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 64e3 } }, "google/gemini-3-pro-preview": { id: "google/gemini-3-pro-preview", name: "Gemini 3 Pro Preview", description: "Preview Gemini flagship for complex reasoning, coding, and rich multimodal prompts", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2025-11-18", last_updated: "2025-11-18", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "SWE-Bench Pro", score: 43.3, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "google/gemini-2.0-flash-lite": { id: "google/gemini-2.0-flash-lite", name: "Gemini 2.0 Flash-Lite", description: "Low-latency Gemini model for high-volume multimodal and agent workloads", family: "gemini-flash-lite", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-06", release_date: "2024-12-11", last_updated: "2024-12-11", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 8192 } }, "google/gemini-3.6-flash": { id: "google/gemini-3.6-flash", name: "Gemini 3.6 Flash", description: "Fast Gemini model balancing multimodal reasoning, tool use, and cost", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-03", release_date: "2026-07-21", last_updated: "2026-07-21", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "SWE-Bench Pro", score: 58.7, metric: "resolve rate", harness: "Antigravity", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "DeepSWE", score: 49, metric: "resolve rate", variant: "high reasoning", version: "1.1", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "Terminal-Bench", score: 78, metric: "accuracy", harness: "Terminus 2", version: "2.1", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "MLE-Bench", score: 63.9, metric: "average position score", dataset: "Partial 30", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "GDPval-AA", score: 1421, metric: "Elo", version: "v2", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "OSWorld-Verified", score: 83, metric: "success rate", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "CharXiv Reasoning", score: 85.2, metric: "accuracy", variant: "no tools", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "CharXiv Reasoning", score: 89.4, metric: "accuracy", variant: "with tools", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "GDM-MRCR", score: 91.8, metric: "accuracy", variant: "128k average, 8-needle", version: "v2", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }, { name: "GDM-MRCR", score: 54, metric: "accuracy", variant: "1M pointwise, 8-needle", version: "v2", source: "https://deepmind.google/models/evals-methodology/gemini-3-6-flash/", date: "2026-07-21" }] }, "google/gemini-3.1-flash-lite": { id: "google/gemini-3.1-flash-lite", name: "Gemini 3.1 Flash Lite", description: "Low-latency Gemini model for high-volume multimodal and agent workloads", family: "gemini-flash-lite", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-05-07", last_updated: "2026-05-07", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemini-3.1-flash-live-preview": { id: "google/gemini-3.1-flash-live-preview", name: "Gemini 3.1 Flash Live Preview", description: "High-quality, low-latency Live API model for real-time dialogue and voice-first AI applications", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: false, temperature: true, knowledge: "2025-01", release_date: "2026-03-26", last_updated: "2026-03-26", modalities: { input: ["text", "image", "video", "audio"], output: ["text", "audio"] }, open_weights: false, limit: { context: 131072, output: 65536 } }, "google/gemini-3.5-flash": { id: "google/gemini-3.5-flash", name: "Gemini 3.5 Flash", description: "Fast Gemini model balancing multimodal reasoning, tool use, and cost", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-05-19", last_updated: "2026-05-19", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "Terminal-Bench", score: 76.2, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "SWE-Bench Pro", score: 55.1, metric: "resolve rate", variant: "single attempt", dataset: "public", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "MCP Atlas", score: 83.6, metric: "success rate", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "Toolathlon", score: 56.5, metric: "success rate", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "OSWorld-Verified", score: 78.4, metric: "success rate", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "MMMU Pro", score: 83.6, metric: "accuracy", variant: "no tools", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "CharXiv Reasoning", score: 84.2, metric: "accuracy", variant: "no tools", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "Humanity's Last Exam", score: 40.2, metric: "accuracy", dataset: "full set, text + MM", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "ARC-AGI-2", score: 72.1, metric: "accuracy", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }, { name: "GDPval-AA", score: 1656, metric: "Elo", source: "https://deepmind.google/models/gemini/flash/", date: "2026-05-19" }] }, "google/veo-3.1-generate-preview": { id: "google/veo-3.1-generate-preview", name: "Veo 3.1 Preview", description: "Video model for prompt-guided generation, editing, and motion workflows", family: "veo", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2025-10-15", last_updated: "2026-01", modalities: { input: ["text", "image"], output: ["video"] }, open_weights: false, limit: { context: 1024, output: 1 } }, "google/gemini-3.1-flash-lite-preview": { id: "google/gemini-3.1-flash-lite-preview", name: "Gemini 3.1 Flash Lite Preview", description: "Low-latency Gemini model for high-volume multimodal and agent workloads", family: "gemini-flash-lite", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-03-03", last_updated: "2026-03-03", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/veo-3.1-fast-generate-preview": { id: "google/veo-3.1-fast-generate-preview", name: "Veo 3.1 Fast Preview", description: "Video model for prompt-guided generation, editing, and motion workflows", family: "veo", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2025-10-15", last_updated: "2026-01-01", modalities: { input: ["text", "image", "video"], output: ["video"] }, open_weights: false, limit: { context: 1024, output: 0 } }, "google/gemma-3-27b-it": { id: "google/gemma-3-27b-it", name: "Gemma 3 27B IT", description: "Largest open Gemma 3 instruction model for multilingual text generation and visual understanding", family: "gemma", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-08", release_date: "2025-03-12", last_updated: "2025-03-12", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/google/gemma-3-27b-it" }] }, "google/gemini-embedding-001": { id: "google/gemini-embedding-001", name: "Gemini Embedding 001", description: "Embedding model for semantic search, retrieval, clustering, and ranking pipelines", family: "gemini", attachment: false, reasoning: false, tool_call: false, temperature: false, knowledge: "2025-05", release_date: "2025-05-20", last_updated: "2025-05-20", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 2048, output: 1 } }, "google/gemini-3.1-flash-image": { id: "google/gemini-3.1-flash-image", name: "Nano Banana 2", description: "Image model for prompt-driven generation, editing, and visual design workflows", family: "gemini-flash", attachment: true, reasoning: true, tool_call: false, temperature: true, knowledge: "2025-01", release_date: "2026-05-28", last_updated: "2026-05-28", modalities: { input: ["text", "image", "video", "pdf"], output: ["text", "image"] }, open_weights: false, limit: { context: 131072, output: 32768 } }, "google/gemini-3.5-flash-lite": { id: "google/gemini-3.5-flash-lite", name: "Gemini 3.5 Flash Lite", description: "Fast Gemini model balancing multimodal reasoning, tool use, and cost", family: "gemini-flash-lite", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-03", release_date: "2026-07-21", last_updated: "2026-07-21", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "SWE-Bench Pro", score: 54.2, metric: "resolve rate", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "Terminal-Bench", score: 54, metric: "accuracy", harness: "Terminus 2", version: "2.1", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "MLE-Bench", score: 39.2, metric: "average position score", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "GDPval-AA", score: 1140, metric: "Elo", version: "v2", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "OSWorld-Verified", score: 74, metric: "success rate", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "CharXiv Reasoning", score: 74.5, metric: "accuracy", variant: "no tools", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "CharXiv Reasoning", score: 76.5, metric: "accuracy", variant: "with tools", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "GDM-MRCR", score: 72.2, metric: "accuracy", variant: "128k average, 8-needle", version: "v2", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }, { name: "GDM-MRCR", score: 21.3, metric: "accuracy", variant: "1M pointwise, 8-needle", version: "v2", source: "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", date: "2026-07-21" }] }, "google/gemini-3-pro-image-preview": { id: "google/gemini-3-pro-image-preview", name: "Nano Banana Pro Preview", description: "Nano Banana Pro for higher-fidelity image generation and design-heavy edits", family: "gemini-pro", attachment: true, reasoning: true, tool_call: false, temperature: true, knowledge: "2025-01", release_date: "2025-11-20", last_updated: "2025-11-20", modalities: { input: ["text", "image"], output: ["text", "image"] }, open_weights: false, limit: { context: 65536, output: 32768 } }, "google/gemini-3.1-flash-tts-preview": { id: "google/gemini-3.1-flash-tts-preview", name: "Gemini 3.1 Flash TTS Preview", description: "Low-latency speech generation with steerable prompts and expressive audio tags", family: "gemini-flash", attachment: false, reasoning: false, tool_call: false, temperature: true, knowledge: "2025-01", release_date: "2026-04-15", last_updated: "2026-04-15", modalities: { input: ["text"], output: ["audio"] }, open_weights: false, limit: { context: 8192, output: 16384 } }, "google/gemini-robotics-er-1.6-preview": { id: "google/gemini-robotics-er-1.6-preview", name: "Gemini Robotics-ER 1.6 Preview", description: "Vision-language model for embodied reasoning: spatial understanding, task planning, and physical-world agentic robotics", family: "gemini", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-04-14", last_updated: "2026-04-14", modalities: { input: ["text", "image", "video", "audio"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 65536 } }, "google/veo-3.1-lite-generate-preview": { id: "google/veo-3.1-lite-generate-preview", name: "Veo 3.1 Lite Preview", description: "Video model for prompt-guided generation, editing, and motion workflows", family: "veo", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2026-03-31", last_updated: "2026-03-31", modalities: { input: ["text", "image"], output: ["video"] }, open_weights: false, limit: { context: 1024, output: 0 } }, "google/gemini-flash-lite-latest": { id: "google/gemini-flash-lite-latest", name: "Gemini Flash-Lite Latest", description: "Fast Gemini model balancing multimodal reasoning, tool use, and cost", family: "gemini-flash-lite", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-03", release_date: "2026-07-21", last_updated: "2026-07-21", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemma-4-31b-it": { id: "google/gemma-4-31b-it", name: "Gemma 4 31B IT", description: "Largest Gemma 4 instruction model for open, self-hosted chat and reasoning", family: "gemma", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-02", last_updated: "2026-04-02", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/google/gemma-4-31B-it" }] }, "google/gemini-embedding-2": { id: "google/gemini-embedding-2", name: "Gemini Embedding 2", description: "Multimodal embedding model mapping text, images, video, audio, and PDFs into a unified embedding space", family: "gemini", attachment: true, reasoning: false, tool_call: false, temperature: false, knowledge: "2025-11", release_date: "2026-04-22", last_updated: "2026-04-22", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 8192, output: 3072 } }, "google/gemini-3-flash-preview": { id: "google/gemini-3-flash-preview", name: "Gemini 3 Flash Preview", description: "New Gemini flash lane bringing frontier-style multimodal reasoning to cheaper runs", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2025-12-17", last_updated: "2025-12-17", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "SWE-Bench Pro", score: 34.63, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }, { name: "SWE-Atlas Codebase QnA", score: 8.2, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 10, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 30.3, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }] }, "google/gemini-3.8-flash": { id: "google/gemini-3.8-flash", name: "Gemini 3.8 Flash", description: "Google's most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-09-02", last_updated: "2026-09-02", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemini-3.5-live-translate-preview": { id: "google/gemini-3.5-live-translate-preview", name: "Gemini 3.5 Live Translate Preview", description: "Low-latency audio-to-audio model for real-time speech translation across 70+ languages", family: "gemini-pro", attachment: false, reasoning: false, tool_call: false, temperature: false, knowledge: "2025-01", release_date: "2026-06-09", last_updated: "2026-06-09", modalities: { input: ["audio"], output: ["audio", "text"] }, open_weights: false, limit: { context: 131072, output: 65536 } }, "google/lyria-3-pro-preview": { id: "google/lyria-3-pro-preview", name: "Lyria 3 Pro Preview", description: "Music generation model for full-length songs from text or images with vocals and structure", family: "lyria", attachment: true, reasoning: false, tool_call: false, structured_output: false, temperature: true, release_date: "2026-03-25", last_updated: "2026-03-25", modalities: { input: ["text", "image"], output: ["text", "audio"] }, open_weights: false, limit: { context: 131072, output: 8192 } }, "google/gemini-3.7-flash": { id: "google/gemini-3.7-flash", name: "Gemini 3.7 Flash", description: "High-efficiency Gemini model for agentic workflows, coding, and multimodal reasoning", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-03", release_date: "2026-08-13", last_updated: "2026-08-13", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "FrontierCode", score: 43.6, metric: "score", version: "1.1 Main", source: "https://deepmind.google/models/model-cards/gemini-3-7-flash/", date: "2026-08-13" }, { name: "DeepSWE", score: 65.3, metric: "resolve rate", version: "1.1", source: "https://deepmind.google/models/model-cards/gemini-3-7-flash/", date: "2026-08-13" }, { name: "Terminal-Bench", score: 85.8, metric: "accuracy", version: "2.1", source: "https://deepmind.google/models/model-cards/gemini-3-7-flash/", date: "2026-08-13" }, { name: "AutomationBench", score: 30.4, metric: "accuracy", dataset: "private set", source: "https://deepmind.google/models/model-cards/gemini-3-7-flash/", date: "2026-08-13" }, { name: "GDP.pdf", score: 34, metric: "accuracy", source: "https://deepmind.google/models/model-cards/gemini-3-7-flash/", date: "2026-08-13" }, { name: "GDM-MRCR", score: 97, metric: "accuracy", variant: "128k average, 8-needle", version: "v2", source: "https://deepmind.google/models/model-cards/gemini-3-7-flash/", date: "2026-08-13" }] }, "google/gemini-2.5-pro": { id: "google/gemini-2.5-pro", name: "Gemini 2.5 Pro", description: "Google's proven reasoning model for coding, math, and multimodal analysis", family: "gemini-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2025-06-17", last_updated: "2025-06-17", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "Aider Polyglot", score: 83.1, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-06-06" }, { name: "Artificial Analysis Coding Index", score: 32, metric: "index", source: "https://openrouter.ai/google/gemini-2.5-pro/benchmarks", date: "2026-06-02" }, { name: "SciCode", score: 42.8, metric: "percent correct", source: "https://openrouter.ai/google/gemini-2.5-pro/benchmarks", date: "2026-06-02" }, { name: "Terminal-Bench Hard", score: 26.5, metric: "success rate", source: "https://openrouter.ai/google/gemini-2.5-pro/benchmarks", date: "2026-06-02" }] }, "google/gemini-flash-latest": { id: "google/gemini-flash-latest", name: "Gemini Flash Latest", description: "High-efficiency Gemini model for agentic workflows, coding, and multimodal reasoning", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-03", release_date: "2026-08-13", last_updated: "2026-08-13", modalities: { input: ["text", "image", "video", "audio", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 } }, "google/gemini-3.1-flash-image-preview": { id: "google/gemini-3.1-flash-image-preview", name: "Nano Banana 2 Preview", description: "Image model for prompt-driven generation, editing, and visual design workflows", family: "gemini-flash", attachment: true, reasoning: true, tool_call: false, temperature: true, knowledge: "2025-01", release_date: "2026-02-26", last_updated: "2026-02-26", modalities: { input: ["text", "image", "pdf"], output: ["text", "image"] }, open_weights: false, limit: { context: 65536, output: 65536 } }, "google/gemini-omni-flash-preview": { id: "google/gemini-omni-flash-preview", name: "Gemini Omni Flash Preview", description: "Video generation and editing model for fast, conversational text- and image-to-video workflows", family: "gemini", attachment: true, reasoning: true, tool_call: false, release_date: "2026-06-30", last_updated: "2026-06-30", modalities: { input: ["text", "image", "video"], output: ["video"] }, open_weights: false, limit: { context: 1048576, output: 57920 }, benchmarks: [{ name: "LMArena Text-to-Video Arena", score: 1527, metric: "Elo", source: "https://venturebeat.com/technology/googles-gemini-omni-flash-hits-the-api-turning-enterprise-video-production-into-a-conversation", date: "2026-06-30" }] }, "google/gemini-2.5-flash": { id: "google/gemini-2.5-flash", name: "Gemini 2.5 Flash", description: "Fast Gemini workhorse for multimodal apps where latency and price matter", family: "gemini-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2025-06-17", last_updated: "2025-06-17", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 65536 }, benchmarks: [{ name: "Aider Polyglot", score: 55.1, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-05-25" }, { name: "Artificial Analysis Coding Index", score: 22.2, metric: "index", source: "https://openrouter.ai/google/gemini-2.5-flash/benchmarks", date: "2026-06-02" }, { name: "SciCode", score: 39.4, metric: "percent correct", source: "https://openrouter.ai/google/gemini-2.5-flash/benchmarks", date: "2026-06-02" }, { name: "Terminal-Bench Hard", score: 13.6, metric: "success rate", source: "https://openrouter.ai/google/gemini-2.5-flash/benchmarks", date: "2026-06-02" }] }, "google/gemma-3-12b-it": { id: "google/gemma-3-12b-it", name: "Gemma 3 12B IT", description: "Open multimodal Gemma instruction model for multilingual text generation and image understanding", family: "gemma", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-08", release_date: "2025-03-12", last_updated: "2025-03-12", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/google/gemma-3-12b-it" }] }, "thinkingmachines/inkling-small": { id: "thinkingmachines/inkling-small", name: "Inkling Small", description: "Multimodal MoE reasoning model (276B total, 12B active) for text, image, and audio", family: "ling", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-07-30", last_updated: "2026-07-30", modalities: { input: ["text", "image", "audio"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 1048576 }, license: "Apache-2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/thinkingmachines/Inkling-Small" }] }, "thinkingmachines/inkling": { id: "thinkingmachines/inkling", name: "Inkling", description: "Multimodal MoE reasoning model (975B total, 41B active) for text, image, and audio", family: "ling", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-07-15", last_updated: "2026-07-15", modalities: { input: ["text", "image", "audio"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 1048576 }, license: "Apache-2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/thinkingmachines/Inkling" }] }, "aisingapore/gemma-sea-lion-v4-27b-it": { id: "aisingapore/gemma-sea-lion-v4-27b-it", name: "Gemma-SEA-LION-v4-27B-IT", description: "Gemma 3 27B tuned by AI Singapore for Southeast Asian languages and instruction following", family: "gemma", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2025-09-23", last_updated: "2025-09-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT" }] }, "meta/llama-3.1-8b-instruct": { id: "meta/llama-3.1-8b-instruct", name: "Llama-3.1-8B-Instruct", description: "Compact open Llama model for lightweight chat, drafting, and self-hosting", family: "llama", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2023-12", release_date: "2024-07-23", last_updated: "2024-07-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct" }] }, "meta/llama-guard-3-8b": { id: "meta/llama-guard-3-8b", name: "Llama-Guard-3-8B", description: "Llama 3.1-based safety classifier for moderating prompts and model responses", family: "llama", attachment: false, reasoning: false, tool_call: false, temperature: true, knowledge: "2023-12", release_date: "2024-07-23", last_updated: "2024-07-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-Guard-3-8B" }] }, "meta/muse-spark-1.3": { id: "meta/muse-spark-1.3", name: "Muse Spark 1.3", description: "Muse Spark 1.3 is a multimodal reasoning model from Meta for long-running agentic, multi-agent, and coding workflows. It improves long-horizon agent collaboration, instruction following, and coding efficiency relative to Muse Spark 1.2.", family: "muse", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-09-02", last_updated: "2026-09-02", modalities: { input: ["text", "image", "video", "pdf", "audio"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 131072 } }, "meta/llama-4-scout-17b-instruct": { id: "meta/llama-4-scout-17b-instruct", name: "Llama 4 Scout 17B Instruct", description: "Open Llama with long-context vision for efficient multimodal agents", family: "llama", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-08", release_date: "2025-04-05", last_updated: "2025-04-05", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 1e7, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct" }] }, "meta/muse-spark-1.2": { id: "meta/muse-spark-1.2", name: "Muse Spark 1.2", description: "Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1 with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows.", family: "muse", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-05", last_updated: "2026-08-05", modalities: { input: ["text", "image", "video", "pdf", "audio"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 131072 } }, "meta/llama-3.2-3b": { id: "meta/llama-3.2-3b", name: "Llama-3.2-3B", description: "Small open Llama base model for lightweight text generation and self-hosting", family: "llama", attachment: false, reasoning: false, tool_call: false, temperature: true, knowledge: "2023-12", release_date: "2024-09-25", last_updated: "2024-09-25", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, license: "Llama 3.2 Community License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-3.2-3B" }] }, "meta/llama-4-maverick-17b-instruct": { id: "meta/llama-4-maverick-17b-instruct", name: "Llama 4 Maverick 17B Instruct", description: "Open multimodal Llama for strong reasoning with efficient everyday serving", family: "llama", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-08", release_date: "2025-04-05", last_updated: "2025-04-05", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct" }], benchmarks: [{ name: "Aider Polyglot", score: 15.6, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-04-06" }, { name: "SWE-Bench Pro", score: 5.24, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "meta/muse-spark-1.1": { id: "meta/muse-spark-1.1", name: "Muse Spark 1.1", description: "Muse Spark is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration.", family: "muse", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-08", last_updated: "2026-07-09", modalities: { input: ["text", "image", "pdf", "video"], output: ["text"] }, open_weights: false, limit: { context: 1048576, output: 131072 }, benchmarks: [{ name: "SWE-Bench Pro", score: 61.5, metric: "resolve rate", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "Terminal-Bench", score: 80, metric: "success rate", version: "2.1", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "DeepSWE", score: 53.3, metric: "resolve rate", version: "1.1", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "MCP Atlas", score: 88.1, metric: "success rate", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "JobBench", score: 54.7, metric: "success rate", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "Toolathlon-Verified", score: 75.6, metric: "success rate", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "Humanity's Last Exam", score: 62.1, metric: "accuracy", variant: "with tools", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "OSWorld-Verified", score: 80.8, metric: "success rate", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "Finance Agent", score: 57.2, metric: "accuracy", version: "v2", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "CharXiv Reasoning", score: 88.4, metric: "accuracy", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }, { name: "BabyVision", score: 76.3, metric: "accuracy", source: "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", date: "2026-07-09" }] }, "meta/muse-glimmer-30b": { id: "meta/muse-glimmer-30b", name: "Muse Glimmer 30B", description: "Muse Glimmer is a 30-billion-parameter open-weight multimodal model from Meta Superintelligence Labs, distilled from Muse Spark for always-on local agents, tool use, coding, and image understanding.", family: "muse", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-01-04", release_date: "2026-08-10", last_updated: "2026-08-10", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, license: "Apache 2.0", links: [{ label: "Announcement", url: "https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model", type: "announcement" }, { label: "Model card", url: "https://huggingface.co/meta-models/Muse-Glimmer-30B", type: "model_card" }, { label: "Developer docs", url: "https://developer.meta.com/ai/models/muse-glimmer/", type: "docs" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-models/Muse-Glimmer-30B" }], benchmarks: [{ name: "MCP Atlas", score: 75.5, metric: "success rate", variant: "public", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "DeepSearch QA", score: 74.6, source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "SWE-Bench Pro", score: 51.2, metric: "resolve rate", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "SWE-Bench Verified", score: 76, metric: "resolve rate", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "Terminal-Bench", score: 51.7, metric: "success rate", variant: "with terminus2", version: "2.1", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "OSWorld-Verified", score: 65.9, metric: "success rate", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "AIME 2026", score: 94.7, metric: "accuracy", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "GPQA Diamond", score: 83.5, metric: "accuracy", variant: "AA", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }, { name: "CharXiv Reasoning", score: 78.8, metric: "accuracy", source: "https://huggingface.co/meta-models/Muse-Glimmer-30B", date: "2026-08-10" }] }, "meta/llama-3.2-1b": { id: "meta/llama-3.2-1b", name: "Llama-3.2-1B", description: "Compact open Llama base model for lightweight and on-device use", family: "llama", attachment: false, reasoning: false, tool_call: false, temperature: true, knowledge: "2023-12", release_date: "2024-09-25", last_updated: "2024-09-25", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, license: "Llama 3.2 Community License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-3.2-1B" }] }, "meta/llama-3.2-11b-vision-instruct": { id: "meta/llama-3.2-11b-vision-instruct", name: "Llama-3.2-11B-Vision-Instruct", description: "Open multimodal Llama model for image understanding, captioning, and visual QA", family: "llama", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2023-12", release_date: "2024-09-25", last_updated: "2024-09-25", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct" }] }, "meta/llama-3.1-70b-instruct": { id: "meta/llama-3.1-70b-instruct", name: "Llama-3.1-70B-Instruct", description: "Open Llama instruction model for multilingual chat, reasoning, and coding", family: "llama", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2023-12", release_date: "2024-07-23", last_updated: "2024-07-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct" }] }, "meta/llama-3.3-70b-instruct": { id: "meta/llama-3.3-70b-instruct", name: "Llama-3.3-70B-Instruct", description: "Popular open Llama workhorse for multilingual chat, coding, and self-hosting", family: "llama", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2023-12", release_date: "2024-12-06", last_updated: "2024-12-06", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 10.7, metric: "index", source: "https://openrouter.ai/meta-llama/llama-3.3-70b-instruct/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 26, metric: "percent correct", source: "https://openrouter.ai/meta-llama/llama-3.3-70b-instruct/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 3, metric: "success rate", source: "https://openrouter.ai/meta-llama/llama-3.3-70b-instruct/benchmarks", date: "2026-03-11" }] }, "bytedance-seed/seed-2.1-pro": { id: "bytedance-seed/seed-2.1-pro", name: "Seed 2.1 Pro", description: "Flagship ByteDance Seed 2.1 model for complex multimodal reasoning, coding, and agents", family: "seed", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-06-23", last_updated: "2026-06-23", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 256e3 } }, "bytedance-seed/seed-1-8": { id: "bytedance-seed/seed-1-8", name: "Seed 1.8", description: "ByteDance Seed model for multimodal reasoning, long-context analysis, and agent workflows", family: "seed", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2025-12-28", last_updated: "2025-12-28", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 64e3 } }, "bytedance-seed/seed-2.0-pro": { id: "bytedance-seed/seed-2.0-pro", name: "Seed 2.0 Pro", description: "Flagship ByteDance Seed 2.0 model for complex multimodal reasoning and long-horizon agent workflows", family: "seed", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-14", last_updated: "2026-02-14", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 128e3 } }, "bytedance-seed/seed-2.0-code": { id: "bytedance-seed/seed-2.0-code", name: "Seed 2.0 Code", description: "ByteDance Seed coding model for multimodal software engineering and long-running agents", family: "seed", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-14", last_updated: "2026-02-14", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 131072 } }, "bytedance-seed/seed-2.1-turbo": { id: "bytedance-seed/seed-2.1-turbo", name: "Seed 2.1 Turbo", description: "Faster ByteDance Seed 2.1 model for multimodal reasoning and latency-sensitive agent workflows", family: "seed", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-06-23", last_updated: "2026-06-23", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 256e3 } }, "bytedance-seed/seed-2.0-mini": { id: "bytedance-seed/seed-2.0-mini", name: "Seed 2.0 Mini", description: "Lightweight ByteDance Seed 2.0 model for low-latency multimodal reasoning and high-volume tasks", family: "seed", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-14", last_updated: "2026-02-14", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 32e3 } }, "bytedance-seed/seed-1-6-vision": { id: "bytedance-seed/seed-1-6-vision", name: "Seed 1.6 Vision", description: "ByteDance Seed multimodal model for image understanding, visual reasoning, and tool-assisted tasks", family: "seed", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2025-08-15", last_updated: "2025-08-15", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 32e3 } }, "bytedance-seed/seed-1-6-flash": { id: "bytedance-seed/seed-1-6-flash", name: "Seed 1.6 Flash", description: "Low-latency ByteDance Seed model for high-throughput chat, extraction, and lightweight tool use", family: "seed", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2025-08-28", last_updated: "2025-08-28", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 32e3 } }, "bytedance-seed/seed-evolving": { id: "bytedance-seed/seed-evolving", name: "Seed Evolving", description: "Rolling ByteDance Seed model for rapidly updated reasoning, coding, and agent capabilities", family: "seed", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-06-23", last_updated: "2026-06-23", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 256e3 } }, "bytedance-seed/seed-1-6": { id: "bytedance-seed/seed-1-6", name: "Seed 1.6", description: "ByteDance Seed model for long-context reasoning, instruction following, and tool-assisted tasks", family: "seed", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2025-10-15", last_updated: "2025-10-15", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 64e3 } }, "bytedance-seed/seed-2.0-lite": { id: "bytedance-seed/seed-2.0-lite", name: "Seed 2.0 Lite", description: "Cost-efficient ByteDance Seed 2.0 model for production chat, analysis, and structured generation", family: "seed", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-02-14", last_updated: "2026-02-14", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 32e3 } }, "bytedance-seed/seed-character": { id: "bytedance-seed/seed-character", name: "Seed Character", description: "ByteDance Seed model optimized for character-driven dialogue and consistent conversational behavior", family: "seed", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-06-23", last_updated: "2026-06-23", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 256e3 } }, "writer/palmyra-x5": { id: "writer/palmyra-x5", name: "Palmyra X5", description: "Reasoning model for deliberate analysis, multi-step problem solving, and tool use", family: "palmyra", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2025-04-28", last_updated: "2025-04-28", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 8192 } }, "writer/palmyra-x4": { id: "writer/palmyra-x4", name: "Palmyra X4", description: "Enterprise language model for workflow automation, coding, data analysis, and tool use", family: "palmyra", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2024-10-09", last_updated: "2025-04-28", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 4096 } }, "zhipuai/glm-4.7": { id: "zhipuai/glm-4.7", name: "GLM-4.7", description: "Mature GLM model for dependable coding, reasoning, and structured agent tasks", family: "glm", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-12-22", last_updated: "2025-12-22", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.7" }], benchmarks: [{ name: "SWE-Bench Verified", score: 73.8, metric: "resolved", source: "https://huggingface.co/zai-org/GLM-4.7" }, { name: "Terminal Bench 2.0", score: 33.4, metric: "score", source: "https://huggingface.co/zai-org/GLM-4.7" }] }, "zhipuai/glm-4.5-air": { id: "zhipuai/glm-4.5-air", name: "GLM-4.5-Air", description: "Lighter GLM-4.5 variant for fast coding assistance and cheaper agents", family: "glm-air", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-07-28", last_updated: "2025-07-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 98304 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.5-Air" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 23.8, metric: "index", source: "https://openrouter.ai/z-ai/glm-4.5-air/benchmarks", date: "2026-05-30" }, { name: "SciCode", score: 30.6, metric: "percent correct", source: "https://openrouter.ai/z-ai/glm-4.5-air/benchmarks", date: "2026-05-30" }, { name: "Terminal-Bench Hard", score: 20.5, metric: "success rate", source: "https://openrouter.ai/z-ai/glm-4.5-air/benchmarks", date: "2026-05-30" }] }, "zhipuai/glm-4.6": { id: "zhipuai/glm-4.6", name: "GLM-4.6", description: "Late GLM-4 workhorse for coding agents, reasoning, and structured tasks", family: "glm", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-09-30", last_updated: "2025-09-30", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.6" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 29.5, metric: "index", source: "https://openrouter.ai/z-ai/glm-4.6/benchmarks", date: "2026-05-22" }, { name: "SciCode", score: 38.4, metric: "percent correct", source: "https://openrouter.ai/z-ai/glm-4.6/benchmarks", date: "2026-05-22" }, { name: "Terminal-Bench Hard", score: 25, metric: "success rate", source: "https://openrouter.ai/z-ai/glm-4.6/benchmarks", date: "2026-05-22" }, { name: "SWE-Bench Pro", score: 9.67, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "zhipuai/glm-4.6v": { id: "zhipuai/glm-4.6v", name: "GLM-4.6V", description: "GLM vision model for visual reasoning, documents, and multimodal agents", family: "glm", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-12-08", last_updated: "2025-12-08", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.6V" }] }, "zhipuai/glm-5.2": { id: "zhipuai/glm-5.2", name: "GLM-5.2", description: "Open flagship GLM for long-horizon coding agents and million-token context work", family: "glm", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-06-13", last_updated: "2026-06-13", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-5.2" }], benchmarks: [{ name: "SWE-Bench Pro", score: 62.1, metric: "resolve rate", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "Terminal-Bench", score: 82.7, metric: "success rate", harness: "Claude Code", version: "2.1", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "FrontierSWE", score: 74.4, metric: "dominance", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "Humanity's Last Exam", score: 40.5, metric: "accuracy", dataset: "text-only subset", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "Humanity's Last Exam", score: 54.7, metric: "accuracy", variant: "with tools", dataset: "text-only subset", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "CritPt", score: 20.9, metric: "accuracy", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "AIME", score: 99.2, metric: "accuracy", version: "2026", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "HMMT", score: 94.4, metric: "accuracy", version: "November 2025", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "HMMT", score: 92.5, metric: "accuracy", version: "February 2026", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "IMOAnswerBench", score: 91, metric: "accuracy", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "GPQA Diamond", score: 91.2, metric: "accuracy", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "NL2Repo", score: 48.9, metric: "resolve rate", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "DeepSWE", score: 46.2, metric: "resolve rate", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "Program Bench", score: 63.7, metric: "score", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "Terminal-Bench", score: 81, metric: "success rate", harness: "Terminus 2", version: "2.1", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "PostTrainBench", score: 34.3, metric: "score", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "SWE Marathon", score: 13, metric: "resolve rate", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "MCP Atlas", score: 76.8, metric: "score", dataset: "public subset", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }, { name: "Tool-Decathlon", score: 48.2, metric: "score", source: "https://z.ai/blog/glm-5.2", date: "2026-06-16" }] }, "zhipuai/glm-4.5-flash": { id: "zhipuai/glm-4.5-flash", name: "GLM-4.5-Flash", description: "Efficient GLM model for fast reasoning, coding, and agent workflows", family: "glm-flash", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-07-28", last_updated: "2025-07-28", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 98304 } }, "zhipuai/glm-5.3-flash": { id: "zhipuai/glm-5.3-flash", name: "GLM-5.3-Flash", description: "Native multimodal GLM model for efficient coding and long-horizon agent tasks", family: "glm-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-26", last_updated: "2026-08-26", modalities: { input: ["text", "image", "video", "pdf"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 131072 } }, "zhipuai/glm-4.5": { id: "zhipuai/glm-4.5", name: "GLM-4.5", description: "Hybrid-reasoning GLM release that made the 4.5 line broadly useful", family: "glm", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-07-28", last_updated: "2025-07-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 98304 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.5" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 26.3, metric: "index", source: "https://openrouter.ai/z-ai/glm-4.5/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 34.8, metric: "percent correct", source: "https://openrouter.ai/z-ai/glm-4.5/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 22, metric: "success rate", source: "https://openrouter.ai/z-ai/glm-4.5/benchmarks", date: "2026-03-11" }] }, "zhipuai/glm-4.5v": { id: "zhipuai/glm-4.5v", name: "GLM-4.5V", description: "GLM vision model for visual reasoning, documents, and multimodal agents", family: "glm", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2025-08-11", last_updated: "2025-08-11", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 64e3, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.5V" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 10.9, metric: "index", source: "https://openrouter.ai/z-ai/glm-4.5v/benchmarks", date: "2026-04-29" }, { name: "SciCode", score: 22.1, metric: "percent correct", source: "https://openrouter.ai/z-ai/glm-4.5v/benchmarks", date: "2026-04-29" }, { name: "Terminal-Bench Hard", score: 5.3, metric: "success rate", source: "https://openrouter.ai/z-ai/glm-4.5v/benchmarks", date: "2026-04-29" }] }, "zhipuai/glm-4.7-flashx": { id: "zhipuai/glm-4.7-flashx", name: "GLM-4.7-FlashX", description: "Efficient GLM model for fast reasoning, coding, and agent workflows", family: "glm-flash", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2026-01-19", last_updated: "2026-01-19", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 2e5, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.7-Flash" }] }, "zhipuai/glm-5": { id: "zhipuai/glm-5", name: "GLM-5", description: "General GLM flagship for coding, analysis, and tool-heavy engineering workflows", family: "glm", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-02-12", last_updated: "2026-02-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 204800, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-5" }], benchmarks: [{ name: "SWE-Bench Verified", score: 72.8, metric: "resolved", source: "https://www.swebench.com/" }, { name: "SWE-Atlas Codebase QnA", score: 20.5, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 24.24, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 28.74, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }] }, "zhipuai/glm-5.1": { id: "zhipuai/glm-5.1", name: "GLM-5.1", description: "Strong GLM coding model for agentic engineering, terminals, and repository generation", family: "glm", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-07", last_updated: "2026-04-07", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 2e5, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-5.1" }], benchmarks: [{ name: "Artificial Analysis Coding Agent Index", score: 52.7, metric: "average pass@1", harness: "Claude Code", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 73.2, metric: "pass@1", harness: "Claude Code", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 19.8, metric: "pass@1", harness: "Claude Code", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 65.1, metric: "pass@1", harness: "Claude Code", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }] }, "zhipuai/glm-5-turbo": { id: "zhipuai/glm-5-turbo", name: "GLM-5-Turbo", description: "Faster GLM-5 lane for coding agents that need lower latency", family: "glm", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-03-16", last_updated: "2026-03-16", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 131072 } }, "zhipuai/glm-5.3": { id: "zhipuai/glm-5.3", name: "GLM-5.3", description: "Flagship GLM model for long-horizon coding, agents, and complex project delivery", family: "glm", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-14", last_updated: "2026-08-14", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 131072 } }, "zhipuai/glm-5v-turbo": { id: "zhipuai/glm-5v-turbo", name: "GLM-5V-Turbo", description: "Fast GLM vision model for screenshots, documents, and multimodal agent tasks", family: "glm", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2026-04-01", last_updated: "2026-04-01", modalities: { input: ["text", "image", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 131072 } }, "zhipuai/glm-4.6v-flash": { id: "zhipuai/glm-4.6v-flash", name: "GLM-4.6V-Flash", description: "Lightweight GLM vision model for visual reasoning, documents, and multimodal agents", family: "glm-flash", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2025-12-08", last_updated: "2025-12-08", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.6V-Flash" }] }, "zhipuai/glm-4.7-flash": { id: "zhipuai/glm-4.7-flash", name: "GLM-4.7-Flash", description: "Budget GLM lane for fast coding help, routing, and everyday automation", family: "glm-flash", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-04", release_date: "2026-01-19", last_updated: "2026-01-19", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 2e5, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/zai-org/GLM-4.7-Flash" }], benchmarks: [{ name: "SWE-Bench Verified", score: 59.2, metric: "resolved", source: "https://huggingface.co/zai-org/GLM-4.7-Flash" }] }, "microsoft/mai-code-1-flash": { id: "microsoft/mai-code-1-flash", name: "MAI-Code-1-Flash", description: "Microsoft coding model built for fast, efficient assistance in everyday developer workflows", family: "mai", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-12", release_date: "2026-06-02", last_updated: "2026-06-08", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 128e3 }, links: [{ label: "Model card", url: "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", type: "model_card" }, { label: "Announcement", url: "https://microsoft.ai/news/introducingmai-code-1-flash/", type: "announcement" }], benchmarks: [{ name: "SWE-Bench Pro", score: 51.2, metric: "resolve rate", harness: "GitHub Copilot", source: "https://microsoft.ai/news/introducingmai-code-1-flash/", date: "2026-06-02" }, { name: "SWE-Bench Verified", score: 71.6, metric: "resolved", source: "https://llm-stats.com/benchmarks/swe-bench-verified" }, { name: "Terminal-Bench", score: 54.8, metric: "success rate", version: "2.0", source: "https://llm-stats.com/benchmarks/terminal-bench-2" }, { name: "GPQA Diamond", score: 84.6, metric: "accuracy", source: "https://llm-stats.com/benchmarks/gpqa" }] }, "microsoft/phi-4-mini": { id: "microsoft/phi-4-mini", name: "Phi-4-mini", description: "Compact Microsoft instruction model tuned for efficient coding assistance, reasoning, and low-latency agent tasks", family: "phi", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2023-10", release_date: "2024-12-11", last_updated: "2024-12-11", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4096 }, links: [{ label: "Weights", url: "https://huggingface.co/microsoft/Phi-4-mini-instruct", type: "weights" }], benchmarks: [{ name: "MMLU", score: 67.3, metric: "accuracy", source: "https://huggingface.co/microsoft/Phi-4-mini-instruct/resolve/main/README.md" }] }, "microsoft/mai-code-1.1-flash": { id: "microsoft/mai-code-1.1-flash", name: "MAI-Code-1.1-Flash", description: "Microsoft coding model with native vision support, optimized for fast and efficient software development", family: "mai", attachment: true, reasoning: true, tool_call: true, structured_output: true, release_date: "2026-08-11", last_updated: "2026-08-11", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 128e3 }, links: [{ label: "Announcement", url: "https://microsoft.ai/news/mai-code-1-1-flash-br-better-faster-at-a-quarter-of-the-cost/", type: "announcement" }] }, "sakana/fugu": { id: "sakana/fugu", name: "Fugu", description: "Multi-agent model for routing expert agents across complex analytical tasks", family: "fugu", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, release_date: "2026-06-15", last_updated: "2026-06-15", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 1e6 }, links: [{ label: "Official model catalog", url: "https://raw.githubusercontent.com/SakanaAI/fugu/refs/heads/main/configs/files/fugu.json", type: "docs" }], benchmarks: [{ name: "SWE Bench Pro", score: 59, source: "https://console.sakana.ai/models" }, { name: "Terminal Bench 2.1", score: 80.2, source: "https://console.sakana.ai/models" }, { name: "LiveCodeBench", score: 92.9, source: "https://console.sakana.ai/models" }, { name: "LiveCodeBench Pro", score: 87.8, source: "https://console.sakana.ai/models" }, { name: "Humanity\u2019s Last Exam", score: 47.2, source: "https://console.sakana.ai/models" }, { name: "CharXiv Reasoning", score: 85.1, source: "https://console.sakana.ai/models" }, { name: "GPQA Diamond", score: 95.5, source: "https://console.sakana.ai/models" }, { name: "SciCode", score: 60.1, source: "https://console.sakana.ai/models" }, { name: "\u03C43 Banking", score: 21.7, source: "https://console.sakana.ai/models" }, { name: "Long Context Reasoning", score: 74.7, source: "https://console.sakana.ai/models" }, { name: "MRCRv2", score: 86.6, source: "https://console.sakana.ai/models" }, { name: "CTI-REALM", score: 67.5, source: "https://console.sakana.ai/models" }] }, "sakana/fugu-ultra": { id: "sakana/fugu-ultra", name: "Fugu Ultra", description: "Quality-first multi-agent model for hard research, analysis, and competitions", family: "fugu", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, release_date: "2026-06-15", last_updated: "2026-06-15", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 1e6 }, links: [{ label: "Official model catalog", url: "https://raw.githubusercontent.com/SakanaAI/fugu/refs/heads/main/configs/files/fugu.json", type: "docs" }], benchmarks: [{ name: "SWE Bench Pro", score: 73.7, source: "https://console.sakana.ai/models" }, { name: "Terminal Bench 2.1", score: 82.1, source: "https://console.sakana.ai/models" }, { name: "LiveCodeBench", score: 93.2, source: "https://console.sakana.ai/models" }, { name: "LiveCodeBench Pro", score: 90.8, source: "https://console.sakana.ai/models" }, { name: "Humanity\u2019s Last Exam", score: 50, source: "https://console.sakana.ai/models" }, { name: "CharXiv Reasoning", score: 86.6, source: "https://console.sakana.ai/models" }, { name: "GPQA Diamond", score: 95.5, source: "https://console.sakana.ai/models" }, { name: "SciCode", score: 58.7, source: "https://console.sakana.ai/models" }, { name: "\u03C43 Banking", score: 20.6, source: "https://console.sakana.ai/models" }, { name: "Long Context Reasoning", score: 73.3, source: "https://console.sakana.ai/models" }, { name: "MRCRv2", score: 93.6, source: "https://console.sakana.ai/models" }, { name: "CTI-REALM", score: 69.4, source: "https://console.sakana.ai/models" }] }, "sakana/sakana-namazu": { id: "sakana/sakana-namazu", name: "Sakana Namazu", description: "Japanese-specialized reasoning model based on Kimi K2.6 and tuned for Japanese language, culture, and business workflows", family: "sakana-namazu", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-03", last_updated: "2026-08-03", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 65536 }, links: [{ label: "Official product page", url: "https://sakana.ai/namazu/", type: "announcement" }, { label: "Official model documentation", url: "https://console.sakana.ai/models?model=sakana-namazu", type: "docs" }], benchmarks: [{ name: "AIME26", score: 96.67, source: "https://console.sakana.ai/models?model=sakana-namazu" }, { name: "MMLU-Pro", score: 90.33, source: "https://console.sakana.ai/models?model=sakana-namazu" }, { name: "LiveCodeBench v6", score: 90.33, source: "https://console.sakana.ai/models?model=sakana-namazu" }, { name: "JFBench", score: 37.4, source: "https://console.sakana.ai/models?model=sakana-namazu" }, { name: "Translation", score: 52.2, source: "https://console.sakana.ai/models?model=sakana-namazu" }, { name: "FairPoliticsQA", score: 56.3, source: "https://console.sakana.ai/models?model=sakana-namazu" }] }, "openbmb/minicpm5-2b": { id: "openbmb/minicpm5-2b", name: "MiniCPM5-2B", description: "Dense 2B-class open-source model for on-device and resource-constrained use, with native long-context support, tool calling, and agentic tasks", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-09-06", last_updated: "2026-09-12", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, license: "apache-2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/openbmb/MiniCPM5-2B" }] }, "openbmb/minicpm5-1b": { id: "openbmb/minicpm5-1b", name: "MiniCPM5-1B", description: "Dense 1B-class open-source model for on-device and resource-constrained use, with native long-context support, Think / No Think chat modes, and tool calling", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-05-19", last_updated: "2026-05-19", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, license: "apache-2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/openbmb/MiniCPM5-1B" }] }, "deepseek/deepseek-v4-flash-0423": { id: "deepseek/deepseek-v4-flash-0423", name: "DeepSeek V4 Flash 0423", description: "Initial DeepSeek V4 Flash snapshot for economical reasoning, coding, and million-token agent workloads", family: "deepseek-flash", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-05", release_date: "2026-04-23", last_updated: "2026-04-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, license: "MIT", weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" }] }, "deepseek/deepseek-v4-flash-vision-exp": { id: "deepseek/deepseek-v4-flash-vision-exp", name: "DeepSeek V4 Flash Vision Exp", description: "Experimental multimodal DeepSeek V4 Flash model for image understanding, coding, and agentic work", family: "deepseek-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-21", last_updated: "2026-08-21", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 384e3 } }, "deepseek/deepseek-v4-pro-0813": { id: "deepseek/deepseek-v4-pro-0813", name: "DeepSeek V4 Pro 0813", description: "DeepSeek V4 Pro snapshot with million-token context and support for thinking and non-thinking modes", family: "deepseek-thinking", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-08-12", last_updated: "2026-08-22", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, license: "MIT", weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813" }] }, "deepseek/deepseek-v4-flash-0731": { id: "deepseek/deepseek-v4-flash-0731", name: "DeepSeek V4 Flash 0731", description: "Official DeepSeek V4 Flash release with enhanced agentic capabilities and integrated DSpark speculative decoding", family: "deepseek-flash", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-05", release_date: "2026-07-31", last_updated: "2026-07-31", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, license: "MIT", weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731" }], benchmarks: [{ name: "Terminal-Bench", score: 82.7, metric: "pass@1", variant: "max", version: "2.1", source: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731" }, { name: "NL2Repo", score: 54.2, metric: "resolve rate", harness: "DeepSeek Harness minimal mode", variant: "max effort", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "CyberGym", score: 76.7, metric: "score", harness: "DeepSeek Harness minimal mode", variant: "max effort", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "DeepSWE", score: 54.4, metric: "resolve rate", harness: "DeepSeek Harness minimal mode", variant: "max effort", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "Toolathlon-Verified", score: 70.3, metric: "score", harness: "DeepSeek Harness minimal mode", variant: "max effort", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "Agents' Last Exam", score: 25.2, metric: "score", harness: "DeepSeek Harness minimal mode", variant: "max effort", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "AutomationBench", score: 25.1, metric: "success rate", variant: "max effort", dataset: "public", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "DSBench-FullStack", score: 68.7, metric: "score", variant: "max effort", dataset: "internal", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }, { name: "DSBench-Hard", score: 59.6, metric: "score", variant: "max effort", dataset: "internal", source: "https://api-docs.deepseek.com/updates/", date: "2026-07-31" }] }, "deepseek/deepseek-v3": { id: "deepseek/deepseek-v3", name: "DeepSeek-V3", description: "Open DeepSeek MoE chat model for coding, math, and general reasoning", family: "deepseek", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2024-12-26", last_updated: "2024-12-26", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, license: "DeepSeek Model License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V3" }] }, "deepseek/deepseek-v4-flash": { id: "deepseek/deepseek-v4-flash", name: "DeepSeek V4 Flash", description: "Fast DeepSeek V4 lane for economical reasoning, coding, and long-context work", family: "deepseek-flash", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-05", release_date: "2026-04-24", last_updated: "2026-04-24", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" }], benchmarks: [{ name: "SWE-Bench Verified", score: 79, metric: "resolved", source: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash" }] }, "deepseek/deepseek-reasoner": { id: "deepseek/deepseek-reasoner", name: "DeepSeek Reasoner", description: "DeepSeek reasoning model for multi-step analysis, math, coding, and tools", family: "deepseek-thinking", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-09", release_date: "2025-12-01", last_updated: "2026-02-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V3.2" }], benchmarks: [{ name: "Aider Polyglot", score: 74.2, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-10-03" }] }, "deepseek/deepseek-v4.1-flash": { id: "deepseek/deepseek-v4.1-flash", name: "DeepSeek V4.1 Flash", description: "DeepSeek V4.1 Flash model for reasoning and agentic coding", family: "deepseek-flash", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-05", release_date: "2026-09-10", last_updated: "2026-09-10", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, license: "MIT" }, "deepseek/deepseek-r1": { id: "deepseek/deepseek-r1", name: "DeepSeek-R1", description: "Classic open reasoning model for transparent math, coding, and deliberate problem solving", family: "deepseek-thinking", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-07", release_date: "2025-01-20", last_updated: "2025-05-29", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-R1" }], benchmarks: [{ name: "Aider Polyglot", score: 56.9, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-01-20" }, { name: "Artificial Analysis Coding Index", score: 15.9, metric: "index", source: "https://openrouter.ai/deepseek/deepseek-r1/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 35.7, metric: "percent correct", source: "https://openrouter.ai/deepseek/deepseek-r1/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 6.1, metric: "success rate", source: "https://openrouter.ai/deepseek/deepseek-r1/benchmarks", date: "2026-03-11" }] }, "deepseek/deepseek-chat": { id: "deepseek/deepseek-chat", name: "DeepSeek Chat", description: "DeepSeek chat model for instruction following, coding, and analysis", family: "deepseek", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-09", release_date: "2025-12-01", last_updated: "2026-02-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V3.2" }], benchmarks: [{ name: "Aider Polyglot", score: 70.2, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-10-03" }] }, "deepseek/deepseek-v3.2": { id: "deepseek/deepseek-v3.2", name: "DeepSeek V3.2", description: "Hybrid-reasoning DeepSeek model with thinking and non-thinking modes, sparse attention, and tool-use", family: "deepseek", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-07", release_date: "2025-12-01", last_updated: "2025-12-01", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 64e3 }, license: "MIT License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V3.2" }] }, "deepseek/deepseek-r1-distill-qwen-32b": { id: "deepseek/deepseek-r1-distill-qwen-32b", name: "DeepSeek-R1-Distill-Qwen-32B", description: "R1 reasoning distilled into Qwen 2.5 32B for efficient open-weight step-by-step problem solving", family: "deepseek-thinking", attachment: false, reasoning: true, tool_call: false, temperature: true, release_date: "2025-01-20", last_updated: "2025-01-20", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B" }] }, "deepseek/deepseek-v4-pro-0423": { id: "deepseek/deepseek-v4-pro-0423", name: "DeepSeek V4 Pro 0423", description: "DeepSeek V4 Pro initial snapshot with million-token context and support for thinking and non-thinking modes", family: "deepseek-thinking", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-05", release_date: "2026-04-23", last_updated: "2026-04-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 } }, "deepseek/deepseek-v3-0324": { id: "deepseek/deepseek-v3-0324", name: "DeepSeek V3 0324", description: "March 2025 checkpoint of DeepSeek-V3 with improved reasoning and coding", family: "deepseek", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2025-03-24", last_updated: "2025-03-24", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 163840, output: 163840 }, weights: [{ label: "Model weights", url: "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324", format: "safetensors" }] }, "deepseek/deepseek-v4-pro": { id: "deepseek/deepseek-v4-pro", name: "DeepSeek V4 Pro", description: "Open MoE flagship with million-token context for coding and long agent runs", family: "deepseek-thinking", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-05", release_date: "2026-04-24", last_updated: "2026-04-24", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1e6, output: 384e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" }], benchmarks: [{ name: "SWE-Bench Verified", score: 80.6, metric: "resolved", source: "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro" }, { name: "Artificial Analysis Coding Agent Index", score: 50.1, metric: "average pass@1", harness: "Claude Code", variant: "high", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 67.8, metric: "pass@1", harness: "Claude Code", variant: "high", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 18, metric: "pass@1", harness: "Claude Code", variant: "high", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 64.7, metric: "pass@1", harness: "Claude Code", variant: "high", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }] }, "deepseek/deepseek-ocr-2": { id: "deepseek/deepseek-ocr-2", name: "DeepSeek OCR 2", description: "High-accuracy OCR model for extracting text from documents, screenshots, receipts, and natural scenes", attachment: true, reasoning: false, tool_call: false, release_date: "2026-01-27", last_updated: "2026-01-27", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 8192, output: 8192 } }, "deepseek/deepseek-v3.1": { id: "deepseek/deepseek-v3.1", name: "DeepSeek-V3.1", description: "Hybrid-reasoning DeepSeek model with thinking and non-thinking modes", family: "deepseek", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-08-21", last_updated: "2025-08-21", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, license: "MIT License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/deepseek-ai/DeepSeek-V3.1" }] }, "amazon/nova-2-lite": { id: "amazon/nova-2-lite", name: "Nova 2 Lite", description: "Multimodal reasoning model for visual analysis, planning, and tool use", family: "nova", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-10", release_date: "2025-12-02", last_updated: "2025-12-01", modalities: { input: ["text", "image", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "amazon/nova-micro": { id: "amazon/nova-micro", name: "Nova Micro", description: "Efficient model for low-latency assistance, extraction, and routine automation", family: "nova-micro", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-10", release_date: "2024-12-03", last_updated: "2024-12-03", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 1e4 } }, "amazon/nova-lite": { id: "amazon/nova-lite", name: "Nova Lite", description: "Efficient model for low-latency assistance, extraction, and routine automation", family: "nova-lite", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-10", release_date: "2024-12-03", last_updated: "2024-12-03", modalities: { input: ["text", "image", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 3e5, output: 1e4 } }, "amazon/nova-premier": { id: "amazon/nova-premier", name: "Nova Premier", description: "Multimodal model for complex analysis, long-context understanding, tool use, and model distillation", family: "nova", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-10", release_date: "2025-04-30", last_updated: "2025-04-30", modalities: { input: ["text", "image", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 1e4 } }, "amazon/nova-pro": { id: "amazon/nova-pro", name: "Nova Pro", description: "Flagship model for demanding analysis, coding, and production agent workflows", family: "nova-pro", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-10", release_date: "2024-12-03", last_updated: "2024-12-03", modalities: { input: ["text", "image", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 3e5, output: 1e4 } }, "trendyol/asure-12b": { id: "trendyol/asure-12b", name: "Trendyol Asure 12B", description: "Turkish-language multimodal instruct model built on Gemma 3 12B for e-commerce text, chat, and image-text tasks", family: "gemma", attachment: true, reasoning: false, tool_call: false, temperature: true, release_date: "2026-02-19", last_updated: "2026-02-20", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072 }, license: "Gemma", weights: [{ label: "Hugging Face", url: "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B" }] }, "inclusionai/ling-3.0-flash-fin": { id: "inclusionai/ling-3.0-flash-fin", name: "Ling 3.0 Flash Fin", description: "Finance-enhanced model for financial research, multi-step investment workflows, and long-horizon planning and execution", family: "ling", attachment: false, reasoning: true, tool_call: true, release_date: "2026-08-27", last_updated: "2026-08-27", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 262144, output: 32768 } }, "openai/gpt-5-nano": { id: "openai/gpt-5-nano", name: "GPT-5 Nano", description: "Tiny GPT-5 lane for routing, extraction, classification, and bulk jobs", family: "gpt-nano", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-05-30", release_date: "2025-08-07", last_updated: "2025-08-07", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-4.1-nano": { id: "openai/gpt-4.1-nano", name: "GPT-4.1 nano", description: "Tiny GPT-4.1 option for classification, routing, and very high-volume tasks", family: "gpt-nano", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-04", release_date: "2025-04-14", last_updated: "2025-04-14", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 1047576, output: 32768 }, benchmarks: [{ name: "Aider Polyglot", score: 8.9, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-04-14" }] }, "openai/gpt-4o-2024-05-13": { id: "openai/gpt-4o-2024-05-13", name: "GPT-4o (2024-05-13)", description: "GPT model for general reasoning, writing, coding, and tool-assisted tasks", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2023-09", release_date: "2024-05-13", last_updated: "2024-05-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 4096 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 24.2, metric: "index", source: "https://openrouter.ai/openai/gpt-4o-2024-05-13/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 30.9, metric: "percent correct", source: "https://openrouter.ai/openai/gpt-4o-2024-05-13/benchmarks", date: "2026-03-11" }] }, "openai/gpt-5-codex": { id: "openai/gpt-5-codex", name: "GPT-5-Codex", description: "Coding-optimized GPT model for repository edits, reviews, and agentic software work", family: "gpt-codex", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-09-15", last_updated: "2025-09-15", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 38.9, metric: "index", source: "https://openrouter.ai/openai/gpt-5-codex/benchmarks", date: "2026-06-01" }, { name: "SciCode", score: 40.9, metric: "percent correct", source: "https://openrouter.ai/openai/gpt-5-codex/benchmarks", date: "2026-06-01" }, { name: "Terminal-Bench Hard", score: 37.9, metric: "success rate", source: "https://openrouter.ai/openai/gpt-5-codex/benchmarks", date: "2026-06-01" }] }, "openai/gpt-5-pro": { id: "openai/gpt-5-pro", name: "GPT-5 Pro", description: "Higher-accuracy GPT-5 tier for tough analysis, coding reviews, and planning", family: "gpt-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-10-06", last_updated: "2025-10-06", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 272e3 } }, "openai/gpt-5.1-codex-mini": { id: "openai/gpt-5.1-codex-mini", name: "GPT-5.1 Codex mini", description: "Coding-optimized GPT model for repository edits, reviews, and agentic software work", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-11-13", last_updated: "2025-11-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-5.1-codex": { id: "openai/gpt-5.1-codex", name: "GPT-5.1 Codex", description: "Codex GPT for repository edits, code review, and practical software agents", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-11-13", last_updated: "2025-11-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-5.6-sol": { id: "openai/gpt-5.6-sol", name: "GPT-5.6 Sol", description: "Frontier GPT-5.6 model for complex professional work, coding, and agentic workflows", family: "gpt-sol", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2026-02-16", release_date: "2026-07-09", last_updated: "2026-07-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 64.6, metric: "resolve rate", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Terminal-Bench", score: 88.8, metric: "success rate", version: "2.1", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "DeepSWE", score: 72.7, metric: "resolve rate", version: "1.1", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "GPQA Diamond", score: 94.6, metric: "accuracy", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "FrontierMath", score: 89, metric: "accuracy", dataset: "Tier 1-3", version: "v2", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "BrowseComp", score: 90.4, metric: "accuracy", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "OSWorld", score: 62.6, metric: "success rate", version: "2.0", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "MMMU Pro", score: 83, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Agents' Last Exam", score: 52.7, source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Toolathlon", score: 58, metric: "success rate", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Artificial Analysis Intelligence Index", score: 58.9, metric: "index score", variant: "max", version: "4.1", source: "https://artificialanalysis.ai/articles/gpt-5-6-has-landed", date: "2026-07-09" }, { name: "Artificial Analysis Coding Agent Index", score: 80, metric: "index score", harness: "Codex", variant: "max", version: "1.1", source: "https://artificialanalysis.ai/articles/gpt-5-6-has-landed", date: "2026-07-09" }] }, "openai/gpt-4o-2024-08-06": { id: "openai/gpt-4o-2024-08-06", name: "GPT-4o (2024-08-06)", description: "GPT model for general reasoning, writing, coding, and tool-assisted tasks", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2023-09", release_date: "2024-08-06", last_updated: "2024-08-06", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 }, benchmarks: [{ name: "Aider Polyglot", score: 23.1, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2024-12-30" }, { name: "Artificial Analysis Coding Index", score: 16.6, metric: "index", source: "https://openrouter.ai/openai/gpt-4o-2024-08-06/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 33.1, metric: "percent correct", source: "https://openrouter.ai/openai/gpt-4o-2024-08-06/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 8.3, metric: "success rate", source: "https://openrouter.ai/openai/gpt-4o-2024-08-06/benchmarks", date: "2026-03-11" }] }, "openai/gpt-5.2-codex": { id: "openai/gpt-5.2-codex", name: "GPT-5.2 Codex", description: "Code-specialist GPT for repository edits, reviews, and long-running software agents", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-08-31", release_date: "2025-12-11", last_updated: "2025-12-11", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 41.04, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "openai/gpt-6-astra": { id: "openai/gpt-6-astra", name: "GPT-6 Astra", description: "GPT-6 Astra is OpenAI's most capable model for complex reasoning, coding, computer use, research, and document creation.", family: "gpt-astra", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2026-04-30", release_date: "2026-09-04", last_updated: "2026-09-04", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "Agents' Last Exam", score: 59.3, source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "OSWorld", score: 72.6, metric: "partial score", dataset: "V2-Offline (v2026.08.08)", version: "2.0", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "ScreenSpot-Pro", score: 92.7, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "AutomationBench", score: 41.4, metric: "success rate", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "BenchCAD", score: 95.9, metric: "geometric overlap", variant: "with tools", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "BrowseComp", score: 91.5, metric: "accuracy", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "Terminal-Bench", score: 57.9, metric: "success rate", version: "4.0", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "DeepSWE", score: 74.1, metric: "resolve rate", version: "1.1", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "FrontierCode", score: 64.5, metric: "score", dataset: "Extended", version: "1.1", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "Terminal-Bench Science", score: 64.6, metric: "success rate", version: "0.1", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "FrontierMath", score: 97.6, metric: "accuracy", dataset: "Tier 4", version: "v2", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "GPQA Diamond", score: 96, metric: "accuracy", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "Humanity's Last Exam", score: 57.2, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "ExploitBench", score: 100, metric: "success rate", variant: "without production safeguards", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "SRE-Bench", score: 88, metric: "success rate", variant: "single attempt", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "ARC-AGI-3", score: 99.9, metric: "RHAE", harness: "Responses API", source: "https://openai.com/index/gpt-6-astra/", date: "2026-09-03" }, { name: "Artificial Analysis Intelligence Index", score: 61, metric: "index score", variant: "max", version: "4.1.1", source: "https://artificialanalysis.ai/articles/benchmarking-gpt-6-astra", date: "2026-09-03" }, { name: "Artificial Analysis Coding Agent Index", score: 67, metric: "index score", harness: "Codex", variant: "max", version: "1.4", source: "https://artificialanalysis.ai/articles/benchmarking-gpt-6-astra", date: "2026-09-03" }] }, "openai/o3-deep-research": { id: "openai/o3-deep-research", name: "o3-deep-research", description: "Research model for long-horizon investigation, synthesis, and analytical reports", family: "o", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2024-05", release_date: "2024-06-26", last_updated: "2024-06-26", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 } }, "openai/gpt-6-astra-fast": { id: "openai/gpt-6-astra-fast", name: "GPT-6 Astra (Fast)", description: "Fast variant of GPT-6 Astra for low-latency assistance and high-volume workloads.", family: "gpt-astra", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, release_date: "2026-09-04", last_updated: "2026-09-04", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 } }, "openai/gpt-5.2-pro": { id: "openai/gpt-5.2-pro", name: "GPT-5.2 Pro", description: "Higher-accuracy GPT-5.2 variant for tougher reasoning and review workflows", family: "gpt-pro", attachment: true, reasoning: true, tool_call: true, structured_output: false, temperature: false, knowledge: "2025-08-31", release_date: "2025-12-11", last_updated: "2025-12-11", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-5.3-codex-spark": { id: "openai/gpt-5.3-codex-spark", name: "GPT-5.3 Codex Spark", description: "Coding-optimized GPT model for repository edits, reviews, and agentic software work", family: "gpt-codex-spark", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-08-31", release_date: "2026-02-05", last_updated: "2026-02-05", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 128e3, input: 1e5, output: 32e3 } }, "openai/gpt-4.1-mini": { id: "openai/gpt-4.1-mini", name: "GPT-4.1 mini", description: "Affordable GPT-4.1 lane for fast coding help and structured extraction", family: "gpt-mini", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-04", release_date: "2025-04-14", last_updated: "2025-04-14", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1047576, output: 32768 }, benchmarks: [{ name: "Aider Polyglot", score: 32.4, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-04-14" }] }, "openai/gpt-5-chat-latest": { id: "openai/gpt-5-chat-latest", name: "GPT-5 Chat (latest)", description: "Chat-tuned GPT model for conversational assistance, writing, and tool workflows", family: "gpt-codex", attachment: true, reasoning: true, tool_call: false, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-08-07", last_updated: "2025-08-07", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-5.4": { id: "openai/gpt-5.4", name: "GPT-5.4", description: "Agent-ready GPT for coding and computer-use workflows at a lower cost", family: "gpt", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-08-31", release_date: "2026-03-05", last_updated: "2026-03-05", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 59.1, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }, { name: "SWE-Atlas Codebase QnA", score: 40.8, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Codebase QnA", score: 36.3, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 44.29, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 44.36, metric: "score", harness: "Codex CLI", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "SWE-Atlas Test Writing", score: 40, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "Artificial Analysis Coding Agent Index", score: 53.6, metric: "average pass@1", harness: "Codex", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 72.4, metric: "pass@1", harness: "Codex", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 18.4, metric: "pass@1", harness: "Codex", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 69.8, metric: "pass@1", harness: "Codex", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Artificial Analysis Coding Agent Index", score: 52.2, metric: "average pass@1", harness: "Cursor CLI", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 72.9, metric: "pass@1", harness: "Cursor CLI", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 18.9, metric: "pass@1", harness: "Cursor CLI", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 64.7, metric: "pass@1", harness: "Cursor CLI", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 75.1, metric: "success rate", version: "2.0", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GPQA Diamond", score: 92.8, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 39.8, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 52.1, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "OSWorld-Verified", score: 75, metric: "success rate", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "BrowseComp", score: 82.7, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GDPval", score: 83, metric: "wins or ties", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "ARC-AGI-2", score: 73.3, metric: "accuracy", variant: "Verified", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 47.6, metric: "accuracy", dataset: "Tier 1-3", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 27.1, metric: "accuracy", dataset: "Tier 4", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "MMMU Pro", score: 81.2, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }] }, "openai/gpt-oss-20b": { id: "openai/gpt-oss-20b", name: "GPT OSS 20B", description: "Open GPT reasoning model for self-hosted agents and controllable deployments", family: "gpt-oss", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2025-08-05", last_updated: "2025-08-05", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/openai/gpt-oss-20b" }] }, "openai/gpt-4-turbo": { id: "openai/gpt-4-turbo", name: "GPT-4 Turbo", description: "Compact GPT model for low-latency assistance and high-volume workloads", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: false, temperature: true, knowledge: "2023-12", release_date: "2023-11-06", last_updated: "2024-04-09", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 4096 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 21.5, metric: "index", source: "https://openrouter.ai/openai/gpt-4-turbo/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 31.9, metric: "percent correct", source: "https://openrouter.ai/openai/gpt-4-turbo/benchmarks", date: "2026-03-11" }] }, "openai/gpt-oss-safeguard-20b": { id: "openai/gpt-oss-safeguard-20b", name: "GPT OSS Safeguard 20B", description: "Safety model for policy screening, moderation, and risk-aware routing workflows", family: "gpt-oss", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2025-10-29", last_updated: "2025-10-29", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/openai/gpt-oss-safeguard-20b" }] }, "openai/gpt-5.1": { id: "openai/gpt-5.1", name: "GPT-5.1", description: "Sharper GPT-5 generation for coding, product work, and tool-assisted tasks", family: "gpt", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-09-30", release_date: "2025-11-13", last_updated: "2025-11-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-5.1-codex-max": { id: "openai/gpt-5.1-codex-max", name: "GPT-5.1 Codex Max", description: "Coding-optimized GPT model for repository edits, reviews, and agentic software work", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-11-13", last_updated: "2025-11-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/o4-mini-deep-research": { id: "openai/o4-mini-deep-research", name: "o4-mini-deep-research", description: "Research model for long-horizon investigation, synthesis, and analytical reports", family: "o-mini", attachment: true, reasoning: true, tool_call: true, temperature: false, knowledge: "2024-05", release_date: "2024-06-26", last_updated: "2024-06-26", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 } }, "openai/gpt-5.1-chat-latest": { id: "openai/gpt-5.1-chat-latest", name: "GPT-5.1 Chat", description: "Chat-tuned GPT-5.1 for polished assistants, writing, and product conversations", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-11-13", last_updated: "2025-11-13", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 } }, "openai/gpt-oss-safeguard-120b": { id: "openai/gpt-oss-safeguard-120b", name: "GPT OSS Safeguard 120B", description: "Safety model for policy screening, moderation, and risk-aware routing workflows", family: "gpt-oss", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2025-10-29", last_updated: "2025-10-29", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/openai/gpt-oss-safeguard-120b" }] }, "openai/o1": { id: "openai/o1", name: "o1", description: "O-series reasoning model for hard analysis, math, coding, and planning", family: "o", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2023-09", release_date: "2024-12-05", last_updated: "2024-12-05", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 }, benchmarks: [{ name: "Aider Polyglot", score: 61.7, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2024-12-21" }] }, "openai/gpt-4o": { id: "openai/gpt-4o", name: "GPT-4o", description: "Omni-era GPT for multimodal chat, practical coding, and general assistants", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2023-09", release_date: "2024-05-13", last_updated: "2024-08-06", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 }, benchmarks: [{ name: "Aider Polyglot", score: 23.1, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2024-12-30" }] }, "openai/gpt-5.6-luna": { id: "openai/gpt-5.6-luna", name: "GPT-5.6 Luna", description: "Cost-efficient GPT-5.6 model for fast, high-volume workloads", family: "gpt-luna", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2026-02-16", release_date: "2026-07-09", last_updated: "2026-07-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 62.7, metric: "resolve rate", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Terminal-Bench", score: 84.7, metric: "success rate", version: "2.1", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "DeepSWE", score: 67.2, metric: "resolve rate", version: "1.1", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "GPQA Diamond", score: 92.3, metric: "accuracy", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "FrontierMath", score: 78.6, metric: "accuracy", dataset: "Tier 1-3", version: "v2", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "BrowseComp", score: 83.3, metric: "accuracy", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "OSWorld", score: 45.6, metric: "success rate", version: "2.0", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "MMMU Pro", score: 78.4, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Agents' Last Exam", score: 50.3, source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Toolathlon", score: 53.4, metric: "success rate", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Artificial Analysis Intelligence Index", score: 51.2, metric: "index score", variant: "max", version: "4.1", source: "https://artificialanalysis.ai/articles/gpt-5-6-has-landed", date: "2026-07-09" }, { name: "Artificial Analysis Coding Agent Index", score: 74.6, metric: "index score", harness: "Codex", variant: "max", version: "1.1", source: "https://artificialanalysis.ai/articles/gpt-5-6-has-landed", date: "2026-07-09" }] }, "openai/whisper-large-v3": { id: "openai/whisper-large-v3", name: "Whisper 3 Large", description: "Open Whisper checkpoint for robust multilingual transcription and captioning", family: "whisper", attachment: false, reasoning: false, tool_call: false, release_date: "2024-10-01", last_updated: "2024-10-01", modalities: { input: ["audio"], output: ["text"] }, open_weights: true, limit: { context: 448, output: 4096 } }, "openai/gpt-5.3-codex": { id: "openai/gpt-5.3-codex", name: "GPT-5.3 Codex", description: "Coding-optimized GPT model for repository edits, reviews, and agentic software work", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-08-31", release_date: "2026-02-05", last_updated: "2026-02-05", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "SWE-Atlas Codebase QnA", score: 32.6, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 42.38, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 38.98, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }] }, "openai/gpt-4o-mini": { id: "openai/gpt-4o-mini", name: "GPT-4o mini", description: "Small omni GPT for cheap multimodal assistance and production-scale traffic", family: "gpt-mini", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2023-09", release_date: "2024-07-18", last_updated: "2024-07-18", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 }, benchmarks: [{ name: "Aider Polyglot", score: 3.6, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2024-12-21" }, { name: "SciCode", score: 22.9, metric: "percent correct", source: "https://openrouter.ai/openai/gpt-4o-mini/benchmarks", date: "2026-03-11" }] }, "openai/gpt-image-1.5": { id: "openai/gpt-image-1.5", name: "GPT-Image-1.5", description: "Image model for prompt-driven generation, editing, and visual design workflows", family: "gpt-image", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2025-11-25", last_updated: "2025-11-25", modalities: { input: ["text", "image"], output: ["text", "image"] }, open_weights: false, limit: { context: 0, output: 0 } }, "openai/o1-pro": { id: "openai/o1-pro", name: "o1-pro", description: "O-series reasoning model for hard analysis, math, coding, and planning", family: "o-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2023-09", release_date: "2025-03-19", last_updated: "2025-03-19", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 } }, "openai/gpt-4.1": { id: "openai/gpt-4.1", name: "GPT-4.1", description: "Long-lived GPT workhorse for coding, instruction following, and production apps", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-04", release_date: "2025-04-14", last_updated: "2025-04-14", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1047576, output: 32768 }, benchmarks: [{ name: "Aider Polyglot", score: 52.4, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-04-14" }] }, "openai/gpt-image-1": { id: "openai/gpt-image-1", name: "GPT-Image-1", description: "OpenAI image model for production generation, edits, and brand-safe visual workflows", family: "gpt-image", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2025-04-24", last_updated: "2025-04-24", modalities: { input: ["text", "image"], output: ["image"] }, open_weights: false, limit: { context: 0, output: 0 } }, "openai/gpt-5.4-nano": { id: "openai/gpt-5.4-nano", name: "GPT-5.4 nano", description: "Cheapest GPT-5.4 lane for simple routing, extraction, and bulk automation", family: "gpt-nano", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-08-31", release_date: "2026-03-17", last_updated: "2026-03-17", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 52.4, metric: "resolve rate", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Terminal-Bench", score: 46.3, metric: "accuracy", variant: "reasoning effort xhigh", version: "2.0", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "MCP Atlas", score: 56.1, metric: "score", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Toolathlon", score: 35.5, metric: "score", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "\u03C4\xB2-Bench Telecom", score: 92.5, metric: "accuracy", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "GPQA Diamond", score: 82.8, metric: "accuracy", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Humanity's Last Exam", score: 37.7, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Humanity's Last Exam", score: 24.3, metric: "accuracy", variant: "without tools", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OSWorld-Verified", score: 39, metric: "success rate", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "MMMU Pro", score: 69.5, metric: "accuracy", variant: "with Python", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "MMMU Pro", score: 66.1, metric: "accuracy", variant: "without tools", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OmniDocBench", score: 0.2419, metric: "overall edit distance", variant: "reasoning effort none", version: "1.5", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OpenAI MRCR", score: 44.2, metric: "accuracy", variant: "8-needle, 64K-128K", version: "v2", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OpenAI MRCR", score: 33.1, metric: "accuracy", variant: "8-needle, 128K-256K", version: "v2", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Graphwalks", score: 73.4, metric: "accuracy", variant: "BFS, 0-128K", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Graphwalks", score: 50.8, metric: "accuracy", variant: "parents, 0-128K", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }] }, "openai/gpt-5.5-pro": { id: "openai/gpt-5.5-pro", name: "GPT-5.5 Pro", description: "Highest-accuracy GPT-5.5 tier for slower, precision-heavy reasoning and coding", family: "gpt-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-12-01", release_date: "2026-04-23", last_updated: "2026-04-23", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "BrowseComp", score: 90.1, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 43.1, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 57.2, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 52.4, metric: "accuracy", dataset: "Tier 1-3", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 39.6, metric: "accuracy", dataset: "Tier 4", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GDPval", score: 82.3, metric: "wins or ties", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GeneBench", score: 33.2, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }] }, "openai/whisper-large-v3-turbo": { id: "openai/whisper-large-v3-turbo", name: "Whisper Large v3 Turbo", description: "Speech transcription model for accurate audio-to-text and captioning workflows", family: "whisper", attachment: false, reasoning: false, tool_call: false, release_date: "2024-10-01", last_updated: "2024-10-01", modalities: { input: ["audio"], output: ["text"] }, open_weights: true, limit: { context: 448, output: 448 } }, "openai/gpt-5.4-mini": { id: "openai/gpt-5.4-mini", name: "GPT-5.4 mini", description: "Strong small GPT for coding subagents, quick tool use, and high-volume work", family: "gpt-mini", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-08-31", release_date: "2026-03-17", last_updated: "2026-03-17", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 54.4, metric: "resolve rate", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Terminal-Bench", score: 60, metric: "accuracy", variant: "reasoning effort xhigh", version: "2.0", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "MCP Atlas", score: 57.7, metric: "score", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Toolathlon", score: 42.9, metric: "score", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "\u03C4\xB2-Bench Telecom", score: 93.4, metric: "accuracy", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "GPQA Diamond", score: 88, metric: "accuracy", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Humanity's Last Exam", score: 41.5, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Humanity's Last Exam", score: 28.2, metric: "accuracy", variant: "without tools", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OSWorld-Verified", score: 72.1, metric: "success rate", variant: "reasoning effort xhigh", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "MMMU Pro", score: 78, metric: "accuracy", variant: "with Python", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "MMMU Pro", score: 76.6, metric: "accuracy", variant: "without tools", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OmniDocBench", score: 0.1263, metric: "overall edit distance", variant: "reasoning effort none", version: "1.5", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OpenAI MRCR", score: 47.7, metric: "accuracy", variant: "8-needle, 64K-128K", version: "v2", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "OpenAI MRCR", score: 33.6, metric: "accuracy", variant: "8-needle, 128K-256K", version: "v2", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Graphwalks", score: 76.3, metric: "accuracy", variant: "BFS, 0-128K", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }, { name: "Graphwalks", score: 71.5, metric: "accuracy", variant: "parents, 0-128K", source: "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", date: "2026-03-17" }] }, "openai/gpt-image-2": { id: "openai/gpt-image-2", name: "GPT-Image-2", description: "Image model for prompt-driven generation, editing, and visual design workflows", family: "gpt-image", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2026-04-21", last_updated: "2026-04-21", modalities: { input: ["text", "image"], output: ["image"] }, open_weights: false, limit: { context: 0, output: 0 } }, "openai/gpt-realtime-whisper": { id: "openai/gpt-realtime-whisper", name: "GPT Realtime Whisper", description: "Streaming speech-to-text model for low-latency transcript deltas from live audio", family: "whisper", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2026-05-07", last_updated: "2026-05-07", modalities: { input: ["audio"], output: ["text"] }, open_weights: false, limit: { context: 0, output: 0 } }, "openai/gpt-3.5-turbo": { id: "openai/gpt-3.5-turbo", name: "GPT-3.5-turbo", description: "Compact GPT model for low-latency assistance and high-volume workloads", family: "gpt", attachment: false, reasoning: false, tool_call: false, structured_output: false, temperature: true, knowledge: "2021-09-01", release_date: "2023-03-01", last_updated: "2023-11-06", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 16385, output: 4096 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 10.7, metric: "index", source: "https://openrouter.ai/openai/gpt-3.5-turbo/benchmarks", date: "2026-03-11" }] }, "openai/gpt-5-mini": { id: "openai/gpt-5-mini", name: "GPT-5 Mini", description: "Small GPT-5 for responsive agents, coding help, and everyday automation", family: "gpt-mini", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-05-30", release_date: "2025-08-07", last_updated: "2025-08-07", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 } }, "openai/gpt-oss-120b": { id: "openai/gpt-oss-120b", name: "GPT OSS 120B", description: "Open GPT reasoning model for self-hosted agents and controllable deployments", family: "gpt-oss", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2025-08-05", last_updated: "2025-08-05", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 32768 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/openai/gpt-oss-120b" }] }, "openai/gpt-5.4-pro": { id: "openai/gpt-5.4-pro", name: "GPT-5.4 Pro", description: "More exact GPT-5.4 tier for demanding professional reasoning and agent tasks", family: "gpt-pro", attachment: true, reasoning: true, tool_call: true, structured_output: false, temperature: false, knowledge: "2025-08-31", release_date: "2026-03-05", last_updated: "2026-03-05", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "GPQA Diamond", score: 94.4, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 42.7, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 58.7, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "BrowseComp", score: 89.3, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GDPval", score: 82, metric: "wins or ties", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 50, metric: "accuracy", dataset: "Tier 1-3", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 38, metric: "accuracy", dataset: "Tier 4", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "ARC-AGI-1", score: 94.5, metric: "accuracy", variant: "Verified", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "ARC-AGI-2", score: 83.3, metric: "accuracy", variant: "Verified", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FinanceAgent", score: 61.5, metric: "accuracy", version: "1.1", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GeneBench", score: 25.6, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }] }, "openai/gpt-5.5-instant": { id: "openai/gpt-5.5-instant", name: "GPT-5.5 Instant", description: "Compact GPT model for low-latency assistance and high-volume workloads", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-12-01", release_date: "2026-05-05", last_updated: "2026-05-28", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 4e5, output: 128e3 } }, "openai/gpt-5.6-terra": { id: "openai/gpt-5.6-terra", name: "GPT-5.6 Terra", description: "Balanced GPT-5.6 model for capable, cost-efficient everyday work", family: "gpt-terra", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2026-02-16", release_date: "2026-07-09", last_updated: "2026-07-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 63.4, metric: "resolve rate", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Terminal-Bench", score: 87.4, metric: "success rate", version: "2.1", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "DeepSWE", score: 69.6, metric: "resolve rate", version: "1.1", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "GPQA Diamond", score: 92.9, metric: "accuracy", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "FrontierMath", score: 84.9, metric: "accuracy", dataset: "Tier 1-3", version: "v2", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "BrowseComp", score: 87.5, metric: "accuracy", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "OSWorld", score: 50.2, metric: "success rate", version: "2.0", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "MMMU Pro", score: 80.7, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Agents' Last Exam", score: 50.4, source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Toolathlon", score: 53.1, metric: "success rate", source: "https://openai.com/index/gpt-5-6/", date: "2026-07-09" }, { name: "Artificial Analysis Intelligence Index", score: 55, metric: "index score", variant: "max", version: "4.1", source: "https://artificialanalysis.ai/articles/gpt-5-6-has-landed", date: "2026-07-09" }, { name: "Artificial Analysis Coding Agent Index", score: 77.4, metric: "index score", harness: "Codex", variant: "max", version: "1.1", source: "https://artificialanalysis.ai/articles/gpt-5-6-has-landed", date: "2026-07-09" }] }, "openai/gpt-4": { id: "openai/gpt-4", name: "GPT-4", description: "GPT model for general reasoning, writing, coding, and tool-assisted tasks", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: false, temperature: true, knowledge: "2023-11", release_date: "2023-11-06", last_updated: "2024-04-09", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 8192, output: 8192 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 13.1, metric: "index", source: "https://openrouter.ai/openai/gpt-4/benchmarks", date: "2026-03-11" }] }, "openai/gpt-5.2": { id: "openai/gpt-5.2", name: "GPT-5.2", description: "Reliable GPT generation for broad coding, writing, and tool-assisted product work", family: "gpt", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-08-31", release_date: "2025-12-11", last_updated: "2025-12-11", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 29.94, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "openai/gpt-5": { id: "openai/gpt-5", name: "GPT-5", description: "Original GPT-5 workhorse for reasoning, coding, writing, and tool workflows", family: "gpt", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-09-30", release_date: "2025-08-07", last_updated: "2025-08-07", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 4e5, input: 272e3, output: 128e3 }, benchmarks: [{ name: "Aider Polyglot", score: 88, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-08-23" }, { name: "SWE-Bench Pro", score: 41.78, metric: "resolve rate", dataset: "public", source: "https://labs.scale.com/leaderboard/swe_bench_pro_public" }] }, "openai/gpt-5.2-chat-latest": { id: "openai/gpt-5.2-chat-latest", name: "GPT-5.2 Chat", description: "Chat-tuned GPT model for conversational assistance, writing, and tool workflows", family: "gpt-codex", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-08-31", release_date: "2025-12-11", last_updated: "2025-12-11", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 } }, "openai/o4-mini": { id: "openai/o4-mini", name: "o4-mini", description: "Fast o-series model for compact reasoning, coding, and tool use", family: "o-mini", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-05", release_date: "2025-04-16", last_updated: "2025-04-16", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 }, benchmarks: [{ name: "Aider Polyglot", score: 72, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-04-16" }] }, "openai/gpt-realtime-2.1": { id: "openai/gpt-realtime-2.1", name: "GPT-Realtime-2.1", description: "Realtime speech-to-speech model with configurable reasoning, tool use, and robust voice-agent behavior", family: "gpt", attachment: true, reasoning: true, tool_call: true, structured_output: false, temperature: false, knowledge: "2024-09-30", release_date: "2026-07-06", last_updated: "2026-07-06", modalities: { input: ["text", "audio", "image"], output: ["text", "audio"] }, open_weights: false, limit: { context: 128e3, input: 96e3, output: 32e3 } }, "openai/o3-mini": { id: "openai/o3-mini", name: "o3-mini", description: "Smaller o-series reasoner for economical coding, math, and planning tasks", family: "o-mini", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-05", release_date: "2024-12-20", last_updated: "2025-01-29", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 }, benchmarks: [{ name: "Aider Polyglot", score: 60.4, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-01-31" }] }, "openai/o3": { id: "openai/o3", name: "o3", description: "Deliberate o-series reasoner for hard math, coding, and multi-step analysis", family: "o", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-05", release_date: "2025-04-16", last_updated: "2025-04-16", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 }, benchmarks: [{ name: "Aider Polyglot", score: 81.3, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-06-25" }] }, "openai/o3-pro": { id: "openai/o3-pro", name: "o3-pro", description: "High-effort o3 tier for difficult technical reasoning and careful answers", family: "o-pro", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2024-05", release_date: "2025-06-10", last_updated: "2025-06-10", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 1e5 }, benchmarks: [{ name: "Aider Polyglot", score: 84.9, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-06-28" }] }, "openai/gpt-5.3-chat-latest": { id: "openai/gpt-5.3-chat-latest", name: "GPT-5.3 Chat (latest)", description: "Chat-tuned GPT model for conversational assistance, writing, and tool workflows", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-08-31", release_date: "2026-03-03", last_updated: "2026-03-03", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 } }, "openai/gpt-5.5": { id: "openai/gpt-5.5", name: "GPT-5.5", description: "Default frontier GPT for coding, computer use, research, and knowledge work", family: "gpt", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-12-01", release_date: "2026-04-23", last_updated: "2026-04-23", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 105e4, input: 922e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Pro", score: 58.6, metric: "resolve rate", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "Terminal-Bench", score: 78.2, metric: "success rate", harness: "Terminus-2", version: "2.1", source: "https://www.anthropic.com/news/claude-opus-4-8", date: "2026-05-28" }, { name: "SWE-Atlas Codebase QnA", score: 45.43, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 44.79, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 42.59, metric: "score", harness: "Codex", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }, { name: "Artificial Analysis Coding Agent Index", score: 65.3, metric: "average pass@1", harness: "Codex", variant: "xhigh", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 80.8, metric: "pass@1", harness: "Codex", variant: "xhigh", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 30.9, metric: "pass@1", harness: "Codex", variant: "xhigh", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 84.1, metric: "pass@1", harness: "Codex", variant: "xhigh", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Artificial Analysis Coding Agent Index", score: 60.4, metric: "average pass@1", harness: "Codex", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 79.1, metric: "pass@1", harness: "Codex", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 26.2, metric: "pass@1", harness: "Codex", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 75.8, metric: "pass@1", harness: "Codex", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Artificial Analysis Coding Agent Index", score: 57.8, metric: "average pass@1", harness: "Cursor CLI", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 75, metric: "pass@1", harness: "Cursor CLI", variant: "medium", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 24.9, metric: "pass@1", harness: "Cursor CLI", variant: "medium", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 73.4, metric: "pass@1", harness: "Cursor CLI", variant: "medium", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 82.7, metric: "success rate", version: "2.0", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GPQA Diamond", score: 93.6, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 41.4, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Humanity's Last Exam", score: 52.2, metric: "accuracy", variant: "with tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "OSWorld-Verified", score: 78.7, metric: "success rate", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "BrowseComp", score: 84.4, metric: "accuracy", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "MMMU Pro", score: 81.2, metric: "accuracy", variant: "no tools", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "ARC-AGI-2", score: 85, metric: "accuracy", variant: "Verified", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 51.7, metric: "accuracy", dataset: "Tier 1-3", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "FrontierMath", score: 35.4, metric: "accuracy", dataset: "Tier 4", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "GDPval", score: 84.9, metric: "wins or ties", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "MCP Atlas", score: 75.3, metric: "success rate", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "Toolathlon", score: 55.6, metric: "success rate", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }, { name: "\u03C4\xB2-Bench Telecom", score: 98, metric: "success rate", variant: "original prompts", source: "https://openai.com/index/introducing-gpt-5-5/", date: "2026-04-23" }] }, "openai/gpt-4o-2024-11-20": { id: "openai/gpt-4o-2024-11-20", name: "GPT-4o (2024-11-20)", description: "GPT model for general reasoning, writing, coding, and tool-assisted tasks", family: "gpt", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2023-09", release_date: "2024-11-20", last_updated: "2024-11-20", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 }, benchmarks: [{ name: "Aider Polyglot", score: 18.2, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2024-12-30" }, { name: "Artificial Analysis Coding Index", score: 16.7, metric: "index", source: "https://openrouter.ai/openai/gpt-4o-2024-11-20/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 33.3, metric: "percent correct", source: "https://openrouter.ai/openai/gpt-4o-2024-11-20/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 8.3, metric: "success rate", source: "https://openrouter.ai/openai/gpt-4o-2024-11-20/benchmarks", date: "2026-03-11" }] }, "sdaia/allam-2-7b": { id: "sdaia/allam-2-7b", name: "ALLaM-2-7b", description: "ALLaM-2-7b instruction tuned model by SDAIA", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2025-01-23", last_updated: "2025-01-23", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 4096, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/ALLaM-AI/ALLaM-2.0-7B-Instruct" }] }, "moonshotai/kimi-k2.7-code-highspeed": { id: "moonshotai/kimi-k2.7-code-highspeed", name: "Kimi K2.7 Code Highspeed", description: "Lower-latency Kimi Code variant for interactive edits and coding-agent loops", family: "kimi-k2", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-01", release_date: "2026-06-12", last_updated: "2026-06-12", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/moonshotai/Kimi-K2.7-Code" }] }, "moonshotai/kimi-k2.6": { id: "moonshotai/kimi-k2.6", name: "Kimi K2.6", description: "Multimodal Kimi workhorse for agent loops, coding tasks, and visual context", family: "kimi-k2", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-01", release_date: "2026-04-21", last_updated: "2026-04-21", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/moonshotai/Kimi-K2.6" }], benchmarks: [{ name: "SWE-Bench Verified", score: 80.2, metric: "resolved", source: "https://huggingface.co/moonshotai/Kimi-K2.6" }, { name: "Artificial Analysis Coding Agent Index", score: 50.5, metric: "average pass@1", harness: "Claude Code", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Atlas Codebase QnA", score: 59.8, metric: "pass@1", harness: "Claude Code", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "SWE-Bench Pro", score: 27.3, metric: "pass@1", harness: "Claude Code", dataset: "hard-aa", source: "https://artificialanalysis.ai/agents/coding-agents" }, { name: "Terminal-Bench", score: 64.3, metric: "pass@1", harness: "Claude Code", version: "2.1", source: "https://artificialanalysis.ai/agents/coding-agents" }] }, "moonshotai/kimi-k2.7-code": { id: "moonshotai/kimi-k2.7-code", name: "Kimi K2.7 Code", description: "Coding-focused Kimi model, stronger on long-horizon repo work with less overthinking", family: "kimi-k2", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-01", release_date: "2026-06-12", last_updated: "2026-06-12", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/moonshotai/Kimi-K2.7-Code" }], benchmarks: [{ name: "Kimi Code Bench", score: 62, harness: "Kimi Code CLI", version: "v2", source: "https://huggingface.co/moonshotai/Kimi-K2.7-Code", date: "2026-06-12" }, { name: "Program Bench", score: 53.6, harness: "Kimi Code CLI", source: "https://huggingface.co/moonshotai/Kimi-K2.7-Code", date: "2026-06-12" }, { name: "MLS Bench Lite", score: 35.1, harness: "Kimi Code CLI", source: "https://huggingface.co/moonshotai/Kimi-K2.7-Code", date: "2026-06-12" }, { name: "MCP Atlas", score: 76, metric: "success rate", harness: "Kimi Code CLI", source: "https://huggingface.co/moonshotai/Kimi-K2.7-Code", date: "2026-06-12" }, { name: "MCP Mark Verified", score: 81.1, metric: "success rate", harness: "Kimi Code CLI", source: "https://huggingface.co/moonshotai/Kimi-K2.7-Code", date: "2026-06-12" }, { name: "Kimi Claw 24/7 Bench", score: 46.9, harness: "Kimi Code CLI", source: "https://huggingface.co/moonshotai/Kimi-K2.7-Code", date: "2026-06-12" }] }, "moonshotai/kimi-k2-thinking": { id: "moonshotai/kimi-k2-thinking", name: "Kimi K2 Thinking", description: "Thinking Kimi model for slower research passes, planning, and hard technical questions", family: "kimi-thinking", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-08", release_date: "2025-11-06", last_updated: "2025-11-06", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/moonshotai/Kimi-K2-Thinking" }], benchmarks: [{ name: "SWE-Bench Verified", score: 71.3, metric: "resolved", source: "https://huggingface.co/moonshotai/Kimi-K2-Thinking" }] }, "moonshotai/kimi-k3": { id: "moonshotai/kimi-k3", name: "Kimi K3", description: "Multimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent work", family: "kimi-k3", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, release_date: "2026-07-16", last_updated: "2026-07-16", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 1048576, output: 131072 }, benchmarks: [{ name: "DeepSWE", score: 67.5, metric: "resolve rate", harness: "Kimi Code", variant: "max effort", version: "1.1", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "Terminal-Bench", score: 88.3, metric: "accuracy", harness: "Kimi Code", variant: "max effort", version: "2.1", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "FrontierSWE", score: 81.2, metric: "dominance score", harness: "Kimi Code", variant: "max effort", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "Program Bench", score: 77.8, metric: "score", harness: "Kimi Code", variant: "max effort", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "SWE Marathon", score: 42, metric: "resolve rate", harness: "Claude Code", variant: "max effort", version: "1.1", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "GDPval-AA", score: 1668, metric: "Elo", variant: "max effort", version: "v2", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "AA-Briefcase", score: 1548, metric: "Elo", variant: "max effort", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "AutomationBench", score: 30.8, metric: "success rate", variant: "max effort", dataset: "600-task public subset", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "JobBench", score: 52.9, metric: "score", variant: "max effort", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "SpreadsheetBench", score: 34.8, metric: "score", harness: "Claude Code", variant: "max effort", version: "2", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "BrowseComp", score: 91.2, metric: "accuracy", variant: "max effort, context compaction", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "CharXiv Reasoning", score: 91.3, metric: "accuracy", variant: "max effort, with tools", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }, { name: "ZeroBench", score: 41, metric: "pass@5", variant: "max effort, with tools", source: "https://www.kimi.com/blog/kimi-k3", date: "2026-07-16" }] }, "moonshotai/kimi-k2.5": { id: "moonshotai/kimi-k2.5", name: "Kimi K2.5", description: "Earlier Kimi frontier model for long-context agents, coding, and multimodal work", family: "kimi-k2", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: false, knowledge: "2025-01", release_date: "2026-01", last_updated: "2026-01", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/moonshotai/Kimi-K2.5" }], benchmarks: [{ name: "SWE-Bench Verified", score: 70.8, metric: "resolved", source: "https://www.swebench.com/" }, { name: "SWE-Atlas Codebase QnA", score: 13.1, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-qna" }, { name: "SWE-Atlas Refactoring", score: 20.95, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-refactoring" }, { name: "SWE-Atlas Test Writing", score: 25.77, metric: "score", harness: "Mini-SWE-Agent", source: "https://labs.scale.com/leaderboard/sweatlas-tw" }] }, "moonshotai/kimi-k2-thinking-turbo": { id: "moonshotai/kimi-k2-thinking-turbo", name: "Kimi K2 Thinking Turbo", description: "Kimi reasoning model for long-horizon research, planning, and tool use", family: "kimi-thinking", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-08", release_date: "2025-11-06", last_updated: "2025-11-06", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/moonshotai/Kimi-K2-Thinking" }] }, "moonshotai/kimi-k2.8-preview": { id: "moonshotai/kimi-k2.8-preview", name: "Kimi K2.8 Preview", description: "Kimi coding model with more efficient thinking and up to 1M context, available through Kimi Code", family: "kimi-k2", attachment: true, reasoning: true, tool_call: true, release_date: "2026-09-11", last_updated: "2026-09-11", modalities: { input: ["text", "image", "video"], output: ["text"] }, open_weights: false, limit: { context: 1048576 } }, "cohere/command-r7b-arabic-02-2025": { id: "cohere/command-r7b-arabic-02-2025", name: "Command R7B Arabic", description: "Open Command R model optimized for Arabic enterprise chat, RAG, and cultural knowledge", family: "command-r", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2025-02-27", last_updated: "2025-02-27", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-r7b-arabic-02-2025" }] }, "cohere/command-a-plus-05-2026": { id: "cohere/command-a-plus-05-2026", name: "Command A Plus", description: "Cohere's stronger command model for multilingual agents and enterprise workflows", family: "command-a", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-04-01", release_date: "2026-05-20", last_updated: "2026-06-09", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 64e3 } }, "cohere/command-a-reasoning-08-2025": { id: "cohere/command-a-reasoning-08-2025", name: "Command A Reasoning", description: "Cohere reasoning model for multilingual enterprise agents, tools, and complex workflows", family: "command-a", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2025-08-21", last_updated: "2025-08-21", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 32e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-a-reasoning-08-2025" }] }, "cohere/command-a-vision-07-2025": { id: "cohere/command-a-vision-07-2025", name: "Command A Vision", description: "Cohere vision model for multilingual document analysis, OCR, and image understanding", family: "command-a", attachment: true, reasoning: false, tool_call: false, temperature: true, knowledge: "2024-06-01", release_date: "2025-07-31", last_updated: "2025-07-31", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-a-vision-07-2025" }] }, "cohere/north-mini-code-1-0": { id: "cohere/north-mini-code-1-0", name: "North Mini Code", description: "Cohere coding model for practical software engineering and agentic edits", family: "north", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2025-09-23", release_date: "2026-06-09", last_updated: "2026-06-09", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 64e3 }, benchmarks: [{ name: "SWE-Bench Verified", score: 67.6, metric: "resolved", harness: "SWE-agent", source: "https://huggingface.co/CohereLabs/North-Mini-Code-1.0", date: "2026-06-09" }, { name: "SWE-Bench Pro", score: 40.2, metric: "resolve rate", harness: "SWE-agent", source: "https://huggingface.co/CohereLabs/North-Mini-Code-1.0", date: "2026-06-09" }, { name: "Artificial Analysis Intelligence Index", score: 27.6, metric: "index score", source: "https://artificialanalysis.ai/articles/north-mini-code-cohere-s-small-coding-focused-moe-model", date: "2026-06-09" }, { name: "Artificial Analysis Coding Index", score: 33.4, metric: "index score", source: "https://artificialanalysis.ai/articles/north-mini-code-cohere-s-small-coding-focused-moe-model", date: "2026-06-09" }, { name: "GDPval-AA", score: 14, metric: "win rate", source: "https://artificialanalysis.ai/articles/north-mini-code-cohere-s-small-coding-focused-moe-model", date: "2026-06-09" }, { name: "\u03C4\xB2-Bench Telecom", score: 37, metric: "success rate", source: "https://artificialanalysis.ai/articles/north-mini-code-cohere-s-small-coding-focused-moe-model", date: "2026-06-09" }] }, "cohere/command-r-plus-08-2024": { id: "cohere/command-r-plus-08-2024", name: "Command R+", description: "Cohere's RAG workhorse for long-context enterprise search and tool use", family: "command-r", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2024-08-30", last_updated: "2024-08-30", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-r-plus-08-2024" }] }, "cohere/command-a-translate-08-2025": { id: "cohere/command-a-translate-08-2025", name: "Command A Translate", description: "Translation model for multilingual conversion, localization, and cross-language workflows", family: "command-a", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2025-08-28", last_updated: "2025-08-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 8e3, output: 8e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-a-translate-08-2025" }] }, "cohere/command-a-03-2025": { id: "cohere/command-a-03-2025", name: "Command A", description: "Cohere command model for multilingual enterprise agents, tools, and chat", family: "command-a", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2025-03-13", last_updated: "2025-03-13", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 8e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-a-03-2025" }], benchmarks: [{ name: "Aider Polyglot", score: 12, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-03-14" }] }, "cohere/c4ai-aya-expanse-32b": { id: "cohere/c4ai-aya-expanse-32b", name: "Aya Expanse 32B", description: "Open multilingual model optimized for generation across 23 languages", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2024-10-24", last_updated: "2024-10-24", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4e3 }, license: "CC-BY-NC-4.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/aya-expanse-32b" }] }, "cohere/c4ai-aya-expanse-8b": { id: "cohere/c4ai-aya-expanse-8b", name: "Aya Expanse 8B", description: "Compact open multilingual model optimized for generation across 23 languages", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2024-10-24", last_updated: "2024-10-24", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 8e3, output: 4e3 }, license: "CC-BY-NC-4.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/aya-expanse-8b" }] }, "cohere/c4ai-aya-vision-8b": { id: "cohere/c4ai-aya-vision-8b", name: "Aya Vision 8B", description: "Compact open multilingual vision model for OCR and visual question answering", attachment: true, reasoning: false, tool_call: false, temperature: true, release_date: "2025-03-04", last_updated: "2025-05-14", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 16e3, output: 4e3 }, license: "CC-BY-NC-4.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/aya-vision-8b" }] }, "cohere/command-r7b-12-2024": { id: "cohere/command-r7b-12-2024", name: "Command R7B", description: "Cohere retrieval model for long-context chat and enterprise RAG workflows", family: "command-r", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2024-12-02", last_updated: "2024-12-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-r7b-12-2024" }] }, "cohere/c4ai-aya-vision-32b": { id: "cohere/c4ai-aya-vision-32b", name: "Aya Vision 32B", description: "Open multilingual vision model for OCR, visual reasoning, and image question answering", attachment: true, reasoning: false, tool_call: false, temperature: true, release_date: "2025-03-04", last_updated: "2025-05-14", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 16e3, output: 4e3 }, license: "CC-BY-NC-4.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/aya-vision-32b" }] }, "cohere/command-r-08-2024": { id: "cohere/command-r-08-2024", name: "Command R", description: "Cohere retrieval model for long-context chat and enterprise RAG workflows", family: "command-r", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-06-01", release_date: "2024-08-30", last_updated: "2024-08-30", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 4e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/CohereLabs/c4ai-command-r-08-2024" }] }, "upstage/solar-pro4": { id: "upstage/solar-pro4", name: "Solar Pro 4", description: "Upstage's flagship model, specialized for agentic use", family: "solar-pro", attachment: false, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-02", release_date: "2026-08-06", last_updated: "2026-08-06", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 524288, output: 131072 } }, "upstage/solar-pro3": { id: "upstage/solar-pro3", name: "Solar Pro 3", description: "Flagship model for demanding analysis, coding, and production agent workflows", family: "solar-pro", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03", release_date: "2026-01", last_updated: "2026-01", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 8192 } }, "upstage/solar-pro2": { id: "upstage/solar-pro2", name: "Solar Pro 2", description: "Flagship model for demanding analysis, coding, and production agent workflows", family: "solar-pro", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-03", release_date: "2025-05-20", last_updated: "2025-05-20", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 65536, output: 8192 } }, "arcee-ai/trinity-mini": { id: "arcee-ai/trinity-mini", name: "Trinity Mini", description: "Reasoning-tuned 26B MoE model with 3B active parameters for agents, tools, and multi-step workloads", family: "trinity", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-12-01", last_updated: "2026-05-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, license: "OpenMDW-1.1", links: [{ label: "Model card", url: "https://huggingface.co/arcee-ai/Trinity-Mini", type: "model_card" }, { label: "Announcement", url: "https://www.arcee.ai/blog/the-trinity-manifesto", type: "announcement" }, { label: "License", url: "https://huggingface.co/arcee-ai/Trinity-Mini/blob/main/LICENSE", type: "license" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/arcee-ai/Trinity-Mini", format: "safetensors" }] }, "arcee-ai/trinity-large-thinking": { id: "arcee-ai/trinity-large-thinking", name: "Trinity Large Thinking", description: "Reasoning-optimized 398B MoE agent model with extended thinking for long-horizon and multi-turn tool use", family: "trinity", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-04-01", last_updated: "2026-05-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 524288, output: 262144 }, license: "OpenMDW-1.1", links: [{ label: "Model card", url: "https://huggingface.co/arcee-ai/Trinity-Large-Thinking", type: "model_card" }, { label: "Announcement", url: "https://www.arcee.ai/blog/trinity-large-thinking", type: "announcement" }, { label: "License", url: "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/blob/main/LICENSE", type: "license" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/arcee-ai/Trinity-Large-Thinking", format: "safetensors" }] }, "arcee-ai/trinity-large-preview": { id: "arcee-ai/trinity-large-preview", name: "Trinity Large Preview", description: "Lightly post-trained 398B MoE chat model for creative work, long-context prompts, and tool-using agents", family: "trinity", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2026-01-27", last_updated: "2026-05-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 524288, output: 262144 }, license: "OpenMDW-1.1", links: [{ label: "Model card", url: "https://huggingface.co/arcee-ai/Trinity-Large-Preview", type: "model_card" }, { label: "Announcement", url: "https://www.arcee.ai/blog/trinity-large", type: "announcement" }, { label: "License", url: "https://huggingface.co/arcee-ai/Trinity-Large-Preview/blob/main/LICENSE", type: "license" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/arcee-ai/Trinity-Large-Preview", format: "safetensors" }] }, "arcee-ai/trinity-nano-preview": { id: "arcee-ai/trinity-nano-preview", name: "Trinity Nano Preview", description: "Experimental chat-tuned 6B MoE model with 1B active parameters for low-resource chat and instruction following", family: "trinity", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2025-12-01", last_updated: "2026-05-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, license: "OpenMDW-1.1", links: [{ label: "Model card", url: "https://huggingface.co/arcee-ai/Trinity-Nano-Preview", type: "model_card" }, { label: "Announcement", url: "https://www.arcee.ai/blog/the-trinity-manifesto", type: "announcement" }, { label: "License", url: "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/blob/main/LICENSE", type: "license" }], weights: [{ label: "Hugging Face", url: "https://huggingface.co/arcee-ai/Trinity-Nano-Preview", format: "safetensors" }] }, "sarvam/sarvam-105b": { id: "sarvam/sarvam-105b", name: "Sarvam 105B", description: "Flagship Indian-language reasoning model for enterprise multilingual applications", family: "sarvam", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-09-01", last_updated: "2025-09-01", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 } }, "sarvam/sarvam-30b": { id: "sarvam/sarvam-30b", name: "Sarvam 30B", description: "Efficient Indian-language reasoning model for chat, coding, and multilingual work", family: "sarvam", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-02-18", last_updated: "2026-02-18", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 } }, "tencent/hy3": { id: "tencent/hy3", name: "Hy3", description: "Tencent Hy reasoning model for coding, instruction following, and agent tasks", family: "Hy", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-07-06", last_updated: "2026-07-06", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, input: 192e3, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/tencent/Hy3" }], benchmarks: [{ name: "SWE-Bench Verified", score: 78, metric: "resolved", source: "https://huggingface.co/tencent/Hy3" }] }, "tencent/hy4-preview": { id: "tencent/hy4-preview", name: "Hy4 preview", description: "A next-generation productivity model with significantly enhanced Agent and complex task execution capabilities.", family: "Hy", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-08-28", last_updated: "2026-08-28", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 1024e3, output: 64e3 } }, "tencent/hy3-preview": { id: "tencent/hy3-preview", name: "Hy3 preview", description: "Tencent Hy reasoning model for coding, instruction following, and agent tasks", family: "Hy", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2026-04-20", last_updated: "2026-04-20", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 64e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/tencent/Hy3-preview" }], benchmarks: [{ name: "SWE-Bench Verified", score: 74.4, metric: "resolved", source: "https://huggingface.co/tencent/Hy3-preview" }] }, "xai/grok-4.3": { id: "xai/grok-4.3", name: "Grok 4.3", description: "xAI's default Grok for chat, coding, agentic tools, and lower hallucination risk", family: "grok", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-17", last_updated: "2026-04-17", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 3e4 }, benchmarks: [{ name: "Artificial Analysis Intelligence Index", score: 53, metric: "index score", version: "4.0", source: "https://artificialanalysis.ai/articles/xai-launches-grok-4-3-with-improved-agentic-performance-and-lower-pricing", date: "2026-04-30" }, { name: "GDPval-AA", score: 1500, metric: "Elo", source: "https://artificialanalysis.ai/articles/xai-launches-grok-4-3-with-improved-agentic-performance-and-lower-pricing", date: "2026-04-30" }, { name: "\u03C4\xB2-Bench Telecom", score: 98, metric: "success rate", source: "https://artificialanalysis.ai/articles/xai-launches-grok-4-3-with-improved-agentic-performance-and-lower-pricing", date: "2026-04-30" }, { name: "IFBench", score: 81, metric: "accuracy", source: "https://artificialanalysis.ai/articles/xai-launches-grok-4-3-with-improved-agentic-performance-and-lower-pricing", date: "2026-04-30" }] }, "xai/grok-4.20-0309-reasoning": { id: "xai/grok-4.20-0309-reasoning", name: "Grok 4.20 (Reasoning)", description: "Reasoning Grok for document-heavy analysis and long-horizon tool use", family: "grok", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-03-09", last_updated: "2026-03-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 3e4 } }, "xai/grok-4.5": { id: "xai/grok-4.5", name: "Grok 4.5", description: "xAI's Grok model for chat, coding, agentic tools, and lower hallucination risk", family: "grok", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-07-08", last_updated: "2026-07-08", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 5e5, output: 5e5 }, benchmarks: [{ name: "SWE-Bench Pro", score: 64.7, metric: "resolve rate", source: "https://x.ai/news/grok-4-5", date: "2026-07-08" }, { name: "SWE-Bench Multilingual", score: 78, metric: "resolve rate", source: "https://x.ai/news/grok-4-5", date: "2026-07-08" }, { name: "Terminal-Bench", score: 83.3, metric: "success rate", version: "2.1", source: "https://x.ai/news/grok-4-5", date: "2026-07-08" }, { name: "DeepSWE", score: 62, metric: "resolve rate", version: "1.0", source: "https://x.ai/news/grok-4-5", date: "2026-07-08" }, { name: "DeepSWE", score: 53, metric: "resolve rate", harness: "mini-swe-agent", version: "1.1", source: "https://x.ai/news/grok-4-5", date: "2026-07-08" }, { name: "SWE Marathon", score: 29, metric: "pass@1", source: "https://x.ai/news/grok-4-5", date: "2026-07-08" }] }, "xai/grok-build-0.1": { id: "xai/grok-build-0.1", name: "Grok Build 0.1", description: "Fast Grok coding model tuned for agentic engineering and iterative edits", family: "grok-build", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-16", last_updated: "2026-04-16", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 256e3, output: 256e3 } }, "xai/grok-4.1-fast-reasoning": { id: "xai/grok-4.1-fast-reasoning", name: "Grok 4.1 Fast (Reasoning)", description: "xAI's fast agentic tool-calling model with a 2M context window and built-in reasoning", family: "grok", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2025-11-19", last_updated: "2025-11-19", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e6, output: 3e4 } }, "xai/grok-imagine-video-1.5": { id: "xai/grok-imagine-video-1.5", name: "Grok Imagine Video 1.5", description: "Video model for image-to-video generation, editing, and extension workflows", family: "grok", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2026-05-30", last_updated: "2026-05-30", modalities: { input: ["text", "image", "video"], output: ["video"] }, open_weights: false, limit: { context: 1024, output: 0 } }, "xai/grok-imagine-image-2.0": { id: "xai/grok-imagine-image-2.0", name: "Grok Imagine Image 2.0", description: "Image model for prompt-driven generation, editing, and visual design workflows", family: "grok", attachment: true, reasoning: false, tool_call: false, temperature: false, release_date: "2026-08-07", last_updated: "2026-08-07", modalities: { input: ["text", "image"], output: ["image"] }, open_weights: false, limit: { context: 8e3, output: 0 } }, "xai/grok-4.6": { id: "xai/grok-4.6", name: "Grok 4.6", description: "xAI's frontier model for long-running agents, coding, knowledge work, and visual projects", family: "grok", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, knowledge: "2026-02-01", release_date: "2026-08-12", last_updated: "2026-08-12", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 5e5, output: 5e5 } }, "xai/grok-4.1-fast": { id: "xai/grok-4.1-fast", name: "Grok 4.1 Fast", description: "xAI's fast agentic tool-calling model with a 2M context window; non-reasoning variant for low-latency responses", family: "grok", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2025-11-19", last_updated: "2025-11-19", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e6, output: 3e4 } }, "xai/grok-4.20-0309-non-reasoning": { id: "xai/grok-4.20-0309-non-reasoning", name: "Grok 4.20 (Non-Reasoning)", description: "Grok model for agentic tool use, reasoning, coding, and live assistance", family: "grok", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2026-03-09", last_updated: "2026-03-09", modalities: { input: ["text", "image", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 3e4 } }, "mistral/pixtral-12b": { id: "mistral/pixtral-12b", name: "Pixtral 12B", description: "Mistral vision-language model for image understanding and multimodal chat", family: "pixtral", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-09", release_date: "2024-09-01", last_updated: "2024-09-01", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Pixtral-12B-2409" }] }, "mistral/codestral-22b-v0.1": { id: "mistral/codestral-22b-v0.1", name: "Codestral-22B-v0.1", description: "Open Mistral code model for fill-in-the-middle and 80+ programming languages", family: "codestral", attachment: false, reasoning: false, tool_call: false, temperature: true, release_date: "2024-05-29", last_updated: "2024-05-29", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 32768, output: 8192 }, license: "Mistral AI Non-Production License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Codestral-22B-v0.1" }] }, "mistral/ministral-3b": { id: "mistral/ministral-3b", name: "Ministral 3B", description: "Compact Mistral model for edge, latency-sensitive, and cost-efficient workloads", family: "ministral", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-03", release_date: "2024-10-16", last_updated: "2024-10-16", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 8192 } }, "mistral/ministral-3-14b-instruct-2512": { id: "mistral/ministral-3-14b-instruct-2512", name: "Ministral 3 14B", description: "Open vision-language model for efficient local deployment, instruction following, and tool use", family: "ministral", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2025-12-02", last_updated: "2025-12-02", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, license: "Apache 2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512" }] }, "mistral/devstral-small-2507": { id: "mistral/devstral-small-2507", name: "Devstral Small", description: "Mistral coding agent model for repository tasks and software engineering workflows", family: "devstral", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-05", release_date: "2025-07-10", last_updated: "2025-07-10", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Devstral-Small-2507" }], benchmarks: [{ name: "SWE-Bench Verified", score: 53.6, metric: "resolved", source: "https://mistral.ai/news/devstral-2507", date: "2025-07-10" }] }, "mistral/mistral-small-2506": { id: "mistral/mistral-small-2506", name: "Mistral Small 3.2", description: "Efficient Mistral model for fast chat, extraction, and production assistants", family: "mistral-small", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-03", release_date: "2025-06-20", last_updated: "2025-06-20", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506" }] }, "mistral/devstral-2512": { id: "mistral/devstral-2512", name: "Devstral 2", description: "Mistral's coding-agent model for repository work, terminal tasks, and software fixes", family: "devstral", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-12", release_date: "2025-12-09", last_updated: "2025-12-09", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 23.7, metric: "index", source: "https://openrouter.ai/mistralai/devstral-2512/benchmarks", date: "2026-05-31" }, { name: "SciCode", score: 33.1, metric: "percent correct", source: "https://openrouter.ai/mistralai/devstral-2512/benchmarks", date: "2026-05-31" }, { name: "Terminal-Bench Hard", score: 18.9, metric: "success rate", source: "https://openrouter.ai/mistralai/devstral-2512/benchmarks", date: "2026-05-31" }] }, "mistral/devstral-small-2": { id: "mistral/devstral-small-2", name: "Devstral Small 2", description: "Compact multimodal coding model for repository exploration, file editing, and software agents", family: "devstral", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-12", release_date: "2025-12-09", last_updated: "2025-12-09", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, license: "Apache-2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512" }] }, "mistral/magistral-medium-latest": { id: "mistral/magistral-medium-latest", name: "Magistral Medium (latest)", description: "Mistral reasoning model for transparent analysis, math, and complex decisions", family: "magistral-medium", attachment: false, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-06", release_date: "2025-03-17", last_updated: "2025-03-20", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 16384 } }, "mistral/mistral-medium-latest": { id: "mistral/mistral-medium-latest", name: "Mistral Medium (latest)", description: "Balanced Mistral model for enterprise assistants, multilingual work, and tools", family: "mistral-medium", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-29", last_updated: "2026-04-29", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 77.6, metric: "resolved", source: "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B" }] }, "mistral/devstral-medium-2507": { id: "mistral/devstral-medium-2507", name: "Devstral Medium", description: "Mistral coding agent model for repository tasks and software engineering workflows", family: "devstral", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-05", release_date: "2025-07-10", last_updated: "2025-07-10", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 128e3 }, benchmarks: [{ name: "SWE-Bench Verified", score: 61.6, metric: "resolved", source: "https://mistral.ai/news/devstral-2507", date: "2025-07-10" }] }, "mistral/mistral-medium-2604": { id: "mistral/mistral-medium-2604", name: "Mistral Medium 3.5", description: "Balanced Mistral model for enterprise assistants, multilingual work, and tools", family: "mistral-medium", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2026-04-29", last_updated: "2026-04-29", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B" }], benchmarks: [{ name: "SWE-Bench Verified", score: 77.6, metric: "resolved", source: "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B" }, { name: "\u03C4\xB3-Telecom", score: 91.4, metric: "accuracy", variant: "public preview", source: "https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5/", date: "2026-05-22" }] }, "mistral/voxtral-mini-3b-2507": { id: "mistral/voxtral-mini-3b-2507", name: "Voxtral Mini 3B 2507", description: "Open audio-language model for speech transcription, audio understanding, and voice-driven tool use", family: "voxtral", attachment: true, reasoning: false, tool_call: true, temperature: true, release_date: "2025-07-15", last_updated: "2025-07-15", modalities: { input: ["text", "audio"], output: ["text"] }, open_weights: true, limit: { context: 32768, output: 32768 }, license: "Apache 2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507" }] }, "mistral/mistral-large-2512": { id: "mistral/mistral-large-2512", name: "Mistral Large 3", description: "Mistral's largest general model for enterprise agents, coding, and multilingual reasoning", family: "mistral-large", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-11", release_date: "2025-12-02", last_updated: "2025-12-02", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 22.7, metric: "index", source: "https://openrouter.ai/mistralai/mistral-large-2512/benchmarks", date: "2026-06-04" }, { name: "SciCode", score: 36.2, metric: "percent correct", source: "https://openrouter.ai/mistralai/mistral-large-2512/benchmarks", date: "2026-06-04" }, { name: "Terminal-Bench Hard", score: 15.9, metric: "success rate", source: "https://openrouter.ai/mistralai/mistral-large-2512/benchmarks", date: "2026-06-04" }] }, "mistral/devstral-medium-latest": { id: "mistral/devstral-medium-latest", name: "Devstral 2 (latest)", description: "Mistral coding agent model for repository tasks and software engineering workflows", family: "devstral", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-12", release_date: "2025-12-02", last_updated: "2025-12-02", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512" }] }, "mistral/voxtral-small-latest": { id: "mistral/voxtral-small-latest", name: "Voxtral Small (latest)", description: "Instruct model with native audio input for speech understanding and tool use", family: "voxtral", attachment: true, reasoning: false, tool_call: true, temperature: true, release_date: "2025-07-15", last_updated: "2025-07-15", modalities: { input: ["text", "audio"], output: ["text"] }, open_weights: true, limit: { context: 32e3, output: 32e3 } }, "mistral/mistral-nemo": { id: "mistral/mistral-nemo", name: "Mistral Nemo", description: "Efficient Mistral-NVIDIA open model for multilingual chat and local deployment", family: "mistral-nemo", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-07", release_date: "2024-07-01", last_updated: "2024-07-01", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407" }] }, "mistral/mistral-small-latest": { id: "mistral/mistral-small-latest", name: "Mistral Small (latest)", description: "Efficient Mistral model for fast chat, extraction, and production assistants", family: "mistral-small", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-06", release_date: "2026-03-16", last_updated: "2026-03-16", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 256e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603" }] }, "mistral/pixtral-large-2502": { id: "mistral/pixtral-large-2502", name: "Pixtral Large (25.02)", description: "Mistral vision-language model for image understanding and multimodal chat", family: "pixtral", attachment: true, reasoning: false, tool_call: true, temperature: true, release_date: "2025-04-08", last_updated: "2025-04-08", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 8192 } }, "mistral/mistral-large-latest": { id: "mistral/mistral-large-latest", name: "Mistral Large (latest)", description: "Flagship Mistral model for advanced reasoning, coding, and multilingual work", family: "mistral-large", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-11", release_date: "2024-11-01", last_updated: "2025-12-02", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512" }] }, "mistral/ministral-3-8b-instruct-2512": { id: "mistral/ministral-3-8b-instruct-2512", name: "Ministral 3 8B", description: "Compact open vision-language model for edge deployment, instruction following, and tool use", family: "ministral", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2025-12-02", last_updated: "2025-12-02", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, license: "Apache 2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512" }] }, "mistral/ministral-3-3b-instruct-2512": { id: "mistral/ministral-3-3b-instruct-2512", name: "Ministral 3 3B", description: "Compact open vision-language model for edge deployment, instruction following, and tool use", family: "ministral", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, release_date: "2025-12-02", last_updated: "2025-12-02", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, license: "Apache 2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512" }] }, "mistral/mistral-large-2411": { id: "mistral/mistral-large-2411", name: "Mistral Large 2.1", description: "Flagship Mistral model for advanced reasoning, coding, and multilingual work", family: "mistral-large", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-11", release_date: "2024-11-18", last_updated: "2024-11-18", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 13.8, metric: "index", source: "https://openrouter.ai/mistralai/mistral-large-2407/benchmarks", date: "2026-03-11" }, { name: "SciCode", score: 29.2, metric: "percent correct", source: "https://openrouter.ai/mistralai/mistral-large-2407/benchmarks", date: "2026-03-11" }, { name: "Terminal-Bench Hard", score: 6.1, metric: "success rate", source: "https://openrouter.ai/mistralai/mistral-large-2407/benchmarks", date: "2026-03-11" }] }, "mistral/ministral-14b": { id: "mistral/ministral-14b", name: "Ministral 14B", description: "Compact multimodal Mistral model for local assistants, edge agents, and efficient tool use", family: "ministral", attachment: true, reasoning: false, tool_call: true, temperature: true, release_date: "2025-12-02", last_updated: "2025-12-02", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 262144, output: 262144 }, license: "Apache-2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512" }] }, "mistral/mistral-small-2603": { id: "mistral/mistral-small-2603", name: "Mistral Small 4", description: "Fast Mistral production model for chat, extraction, and cost-sensitive agents", family: "mistral-small", attachment: true, reasoning: true, tool_call: true, temperature: true, knowledge: "2025-06", release_date: "2026-03-16", last_updated: "2026-03-16", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 256e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603" }], benchmarks: [{ name: "Artificial Analysis Coding Index", score: 24.3, metric: "index", source: "https://openrouter.ai/mistralai/mistral-small-2603/benchmarks", date: "2026-06-01" }, { name: "SciCode", score: 38, metric: "percent correct", source: "https://openrouter.ai/mistralai/mistral-small-2603/benchmarks", date: "2026-06-01" }, { name: "Terminal-Bench Hard", score: 17.4, metric: "success rate", source: "https://openrouter.ai/mistralai/mistral-small-2603/benchmarks", date: "2026-06-01" }] }, "mistral/mistral-small-3-1-24b-instruct-2503": { id: "mistral/mistral-small-3-1-24b-instruct-2503", name: "Mistral Small 3.1 24B", description: "Efficient multimodal model for instruction following, coding, reasoning, and function calling", family: "mistral-small", attachment: true, reasoning: false, tool_call: true, structured_output: true, temperature: true, knowledge: "2024-06", release_date: "2025-03-17", last_updated: "2025-03-17", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 16384 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503" }] }, "mistral/mistral-medium-2505": { id: "mistral/mistral-medium-2505", name: "Mistral Medium 3", description: "Mistral model for multilingual chat, reasoning, and tool-assisted workflows", family: "mistral-medium", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2025-05", release_date: "2025-05-07", last_updated: "2025-05-07", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 131072, output: 131072 }, benchmarks: [{ name: "Artificial Analysis Coding Index", score: 13.6, metric: "index", source: "https://openrouter.ai/mistralai/mistral-medium-3/benchmarks", date: "2026-05-30" }, { name: "SciCode", score: 33.1, metric: "percent correct", source: "https://openrouter.ai/mistralai/mistral-medium-3/benchmarks", date: "2026-05-30" }, { name: "Terminal-Bench Hard", score: 3.8, metric: "success rate", source: "https://openrouter.ai/mistralai/mistral-medium-3/benchmarks", date: "2026-05-30" }] }, "mistral/voxtral-small-24b-2507": { id: "mistral/voxtral-small-24b-2507", name: "Voxtral Small 24B 2507", description: "Open audio-language model for speech transcription, audio understanding, and voice-driven tool use", family: "voxtral", attachment: true, reasoning: false, tool_call: true, temperature: true, release_date: "2025-07-15", last_updated: "2025-07-15", modalities: { input: ["text", "audio"], output: ["text"] }, open_weights: true, limit: { context: 32768, output: 32768 }, license: "Apache 2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Voxtral-Small-24B-2507" }] }, "mistral/codestral-latest": { id: "mistral/codestral-latest", name: "Codestral (latest)", description: "Mistral code model for completions, refactors, and developer IDE workflows", family: "codestral", attachment: false, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-10", release_date: "2024-05-29", last_updated: "2025-01-04", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 256e3, output: 4096 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Codestral-22B-v0.1" }], benchmarks: [{ name: "Aider Polyglot", score: 11.1, metric: "percent correct", source: "https://aider.chat/docs/leaderboards/", date: "2025-01-13" }] }, "mistral/magistral-small-2509": { id: "mistral/magistral-small-2509", name: "Magistral Small 1.2", description: "Open multimodal reasoning model for transparent analysis of text and images", family: "magistral", attachment: true, reasoning: true, tool_call: true, temperature: true, release_date: "2025-09-18", last_updated: "2025-09-18", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 131072 }, license: "Apache 2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Magistral-Small-2509" }] }, "mistral/magistral-small-2506": { id: "mistral/magistral-small-2506", name: "Magistral Small", description: "Open Mistral reasoning model for transparent step-by-step problem solving", family: "magistral", attachment: false, reasoning: true, tool_call: true, temperature: true, release_date: "2025-06-10", last_updated: "2025-06-10", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, license: "Apache 2.0", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Magistral-Small-2506" }] }, "mistral/ministral-8b-instruct-2410": { id: "mistral/ministral-8b-instruct-2410", name: "Ministral 8B Instruct", description: "Efficient open Mistral edge model for on-device chat and function calling", family: "ministral", attachment: false, reasoning: false, tool_call: true, temperature: true, release_date: "2024-10-16", last_updated: "2024-10-16", modalities: { input: ["text"], output: ["text"] }, open_weights: true, limit: { context: 131072, output: 8192 }, license: "Mistral Research License", weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Ministral-8B-Instruct-2410" }] }, "mistral/pixtral-large-latest": { id: "mistral/pixtral-large-latest", name: "Pixtral Large (latest)", description: "Mistral's larger vision model for document-heavy image understanding and chat", family: "pixtral", attachment: true, reasoning: false, tool_call: true, temperature: true, knowledge: "2024-11", release_date: "2024-11-01", last_updated: "2024-11-04", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: true, limit: { context: 128e3, output: 128e3 }, weights: [{ label: "Hugging Face", url: "https://huggingface.co/mistralai/Pixtral-Large-Instruct-2411" }] }, "perplexity/sonar": { id: "perplexity/sonar", name: "Sonar", description: "Fast web-grounded Sonar for current answers, citations, and lightweight retrieval", family: "sonar", attachment: false, reasoning: false, tool_call: false, temperature: true, knowledge: "2025-09-01", release_date: "2024-01-01", last_updated: "2025-09-01", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 4096 }, benchmarks: [{ name: "SciCode", score: 22.9, metric: "percent correct", source: "https://openrouter.ai/perplexity/sonar/benchmarks", date: "2026-03-11" }] }, "perplexity/sonar-reasoning-pro": { id: "perplexity/sonar-reasoning-pro", name: "Sonar Reasoning Pro", description: "Web-grounded Sonar for multi-step research questions that need cited reasoning", family: "sonar-reasoning", attachment: true, reasoning: true, tool_call: false, temperature: true, knowledge: "2025-09-01", release_date: "2024-01-01", last_updated: "2025-09-01", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 4096 } }, "perplexity/sonar-pro": { id: "perplexity/sonar-pro", name: "Sonar Pro", description: "Deeper Sonar search model with broader retrieval and stronger synthesis", family: "sonar-pro", attachment: true, reasoning: false, tool_call: false, temperature: true, knowledge: "2025-09-01", release_date: "2024-01-01", last_updated: "2025-09-01", modalities: { input: ["text", "image"], output: ["text"] }, open_weights: false, limit: { context: 2e5, output: 8192 }, benchmarks: [{ name: "SciCode", score: 22.6, metric: "percent correct", source: "https://openrouter.ai/perplexity/sonar-pro/benchmarks", date: "2026-03-11" }] }, "perplexity/sonar-deep-research": { id: "perplexity/sonar-deep-research", name: "Sonar Deep Research", description: "Sonar search model for autonomous research and citation-backed long-form reports", family: "sonar", attachment: false, reasoning: true, tool_call: false, temperature: false, knowledge: "2025-01", release_date: "2025-02-01", last_updated: "2025-09-01", modalities: { input: ["text"], output: ["text"] }, open_weights: false, limit: { context: 128e3, output: 32768 } }, "vispark/vision-large": { id: "vispark/vision-large", name: "Vision Large", description: "Most capable Vision model for complex reasoning, detailed media analysis, and structured output over a 1M-token context window.", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2024-05-15", last_updated: "2026-09", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "vispark/vision-medium": { id: "vispark/vision-medium", name: "Vision Medium", description: "Balanced multimodal model pairing a 1M-token context window with deeper reasoning for analysis, content creation, and tool use across text, image, audio, video, and PDF inputs.", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2024-05-15", last_updated: "2026-09", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } }, "vispark/vision-small": { id: "vispark/vision-small", name: "Vision Small", description: "Fast, low-cost multimodal model for understanding text, images, audio, video, and PDFs, with tool calling and a 1M-token context window.", attachment: true, reasoning: true, tool_call: true, structured_output: true, temperature: true, release_date: "2024-05-15", last_updated: "2026-09", modalities: { input: ["text", "image", "audio", "video", "pdf"], output: ["text"] }, open_weights: false, limit: { context: 1e6, output: 65536 } } } };
 
 // src/registry.ts
 var REGISTRY_URL = "https://models.dev/models.json";
@@ -49984,6 +49992,9 @@ async function contextFromRegistry(model, host) {
 function peekRegistryContext(model, host) {
   return registryLookup(cache2, model, host);
 }
+function peekRegistryOutputLimit(model, host) {
+  return registryLookup(cache2, model, host, "output");
+}
 var VARIANT_SUFFIXES = [
   "-interleaved-thinking",
   "-thinking",
@@ -50011,7 +50022,7 @@ function modelVariants(name) {
   }
   return variants;
 }
-function registryLookup(reg, model, host) {
+function registryLookup(reg, model, host, field = "context") {
   if (!reg || !model) return void 0;
   const provider = host ? providerFromHost(host) : void 0;
   const roots = [model];
@@ -50027,8 +50038,8 @@ function registryLookup(reg, model, host) {
     const candidates = provider ? [`${provider}/${name}`, name] : [name];
     for (const key of candidates) {
       const entry = reg[key];
-      const ctx = entry?.limit?.context;
-      if (typeof ctx === "number" && ctx > 0) return ctx;
+      const value = entry?.limit?.[field];
+      if (typeof value === "number" && value > 0) return value;
     }
     if (provider === void 0) {
       const suffix = `/${name}`;
@@ -50037,16 +50048,16 @@ function registryLookup(reg, model, host) {
       const parts = [];
       for (const key of Object.keys(reg)) {
         if (!key.endsWith(suffix)) continue;
-        const ctx = reg[key].limit?.context;
-        if (typeof ctx !== "number" || ctx <= 0) continue;
-        if (max === void 0 || ctx > max) max = ctx;
-        distinct.add(ctx);
-        parts.push(`${key}=${ctx}`);
+        const value = reg[key].limit?.[field];
+        if (typeof value !== "number" || value <= 0) continue;
+        if (max === void 0 || value > max) max = value;
+        distinct.add(value);
+        parts.push(`${key}=${value}`);
       }
       if (max !== void 0) {
         if (distinct.size > 1 && !warnedConflicts.has(name)) {
           warnedConflicts.add(name);
-          log("warn", `[acp-registry] conflicting context windows for "${name}" (${parts.join(", ")}) \u2014 using max ${max}`);
+          log("warn", `[acp-registry] conflicting ${field === "context" ? "context windows" : "output ceilings"} for "${name}" (${parts.join(", ")}) \u2014 using max ${max}`);
         }
         return max;
       }
@@ -52093,6 +52104,9 @@ function buildSession(parsed) {
       // guard existed) which would otherwise revive after upgrade and
       // feed the /acp panel + web stats as negative percentages.
       lastInputTokens: Math.max(0, stats.lastInputTokens ?? parsed.lastInputTokens ?? 0),
+      // #857: provenance — legacy files lack it; absent stays absent and
+      // evidence-grade consumers treat absent as untrusted.
+      lastInputTokensSource: stats.lastInputTokensSource === "usage" || stats.lastInputTokensSource === "estimate" ? stats.lastInputTokensSource : void 0,
       // In-memory only — a fresh process has no pending compress fold.
       compressCreditTokens: 0,
       contextTokens: Math.max(0, stats.contextTokens ?? parsed.contextTokens ?? 0)
@@ -52227,6 +52241,14 @@ function lastCompressSuffix(info) {
   if (!info) return "";
   return ` [after compress: shrink ${Math.round(info.shrinkRatio * 100)}% foldPoint=${info.foldPoint} blocks=${info.blocks} ~${info.tokensCompressed}tok]`;
 }
+function storeEffectiveConfig(session, config) {
+  session.metadata["effectiveConfig"] = config;
+}
+function effectiveConfig(session, fallback) {
+  const stored = session?.metadata["effectiveConfig"];
+  if (stored && typeof stored === "object") return { ...fallback, ...stored };
+  return fallback;
+}
 var sessions = /* @__PURE__ */ new Map();
 var MAX_SESSIONS = Math.max(1, Number.parseInt(process.env.BILI_MAX_SESSIONS ?? "256", 10) || 256);
 var initialized = false;
@@ -52352,6 +52374,7 @@ function resetSessionCompression(session) {
   session.state = createInitialState();
   session.blockContents.clear();
   session.stats.lastInputTokens = 0;
+  delete session.stats.lastInputTokensSource;
   session.stats.contextTokens = 0;
   session.metadata.nativeCompactionAt = Date.now();
   markDirty(session);
@@ -52691,12 +52714,20 @@ Use the read tool to access it.`;
     } catch (e) {
       return `${header}
 [Failed to write to ${outPath}: ${String(e)}]
-${body.slice(0, 4e3)}...`;
+${safePrefix(body, 4e3)}...`;
     }
   }
   return `${header}
 ${body}`;
 }
+function safePrefix(text, n) {
+  let cut = Math.min(n, text.length);
+  if (cut > 0 && cut < text.length) {
+    const c = text.charCodeAt(cut - 1);
+    if (c >= 55296 && c <= 56319) cut -= 1;
+  }
+  return text.slice(0, cut);
+}
 function executeSearchContext(args, core, state) {
   const query = typeof args.query === "string" ? args.query : "";
   if (query.length === 0) return "[search_context FAILED: query is required]";
@@ -52708,7 +52739,7 @@ function executeSearchContext(args, core, state) {
   }
   const lines = blocks.map((b2) => {
     const topic = b2.topic ?? "(no topic)";
-    const preview = b2.summary.length > 200 ? b2.summary.slice(0, 200) + "..." : b2.summary;
+    const preview = b2.summary.length > 200 ? safePrefix(b2.summary, 200) + "..." : b2.summary;
     return `${b2.blockId} (T${b2.tier}) "${topic}"
   ${preview}`;
   });
@@ -53456,7 +53487,7 @@ var MAX_PREFLIGHT_ROUNDS = 8;
 var CHUNK_FRACTION = 0.6;
 var MIN_CHUNK_TOKENS = 2e3;
 var MIN_SUMMARY_CHARS = 50;
-var MAX_SUMMARY_OUTPUT_TOKENS = 8192;
+var MAX_SUMMARY_OUTPUT_TOKENS = 32768;
 var MAX_SUMMARY_CALLS_PER_PREFLIGHT = 8;
 function refMaps(messages, state) {
   const refToIdx = /* @__PURE__ */ new Map();
@@ -53551,15 +53582,27 @@ function splitChunks(messages, startIdx, endIdx, budget, minUnits, countText = d
   }
   return chunks;
 }
-function summaryPayload(protocol, model, system, content, stream2, includeMaxOutputTokens) {
+function safeHost(url) {
+  try {
+    return new URL(url).host;
+  } catch {
+    return void 0;
+  }
+}
+function summaryOutputTokens(model, host) {
+  const known = peekRegistryOutputLimit(model, host);
+  return known === void 0 ? MAX_SUMMARY_OUTPUT_TOKENS : Math.min(MAX_SUMMARY_OUTPUT_TOKENS, known);
+}
+function summaryPayload(protocol, model, system, content, stream2, includeMaxOutputTokens, host) {
+  const maxOutputTokens = summaryOutputTokens(model, host);
   if (protocol === "anthropic") {
-    return { model, max_tokens: MAX_SUMMARY_OUTPUT_TOKENS, system, messages: [{ role: "user", content }], stream: stream2 };
+    return { model, max_tokens: maxOutputTokens, system, messages: [{ role: "user", content }], stream: stream2 };
   }
   if (protocol === "openai") {
-    return { model, max_tokens: MAX_SUMMARY_OUTPUT_TOKENS, messages: [{ role: "system", content: system }, { role: "user", content }], stream: stream2 };
+    return { model, max_tokens: maxOutputTokens, messages: [{ role: "system", content: system }, { role: "user", content }], stream: stream2 };
   }
   const payload = { model, instructions: system, input: [{ role: "user", content }], stream: stream2, store: false };
-  if (includeMaxOutputTokens) payload.max_output_tokens = MAX_SUMMARY_OUTPUT_TOKENS;
+  if (includeMaxOutputTokens) payload.max_output_tokens = maxOutputTokens;
   return payload;
 }
 var STREAM_REQUIRED_RE = /\bstream\b[^\n]{0,60}\btrue\b/i;
@@ -53848,7 +53891,7 @@ async function requestSummaryBody(deps, body) {
   }
 }
 async function requestSummary(deps, system, content, stream2, includeMaxOutputTokens) {
-  const text = await requestSummaryBody(deps, JSON.stringify(summaryPayload(deps.protocol, deps.model, system, content, stream2, includeMaxOutputTokens)));
+  const text = await requestSummaryBody(deps, JSON.stringify(summaryPayload(deps.protocol, deps.model, system, content, stream2, includeMaxOutputTokens, safeHost(deps.url))));
   let json;
   try {
     json = JSON.parse(text);
@@ -54050,7 +54093,13 @@ async function preflightCompress(deps, messages) {
       failure = { kind: "exhausted", detail: `the compress budget was exhausted after ${MAX_PREFLIGHT_ROUNDS} rounds${unusableNote}` };
     }
   }
-  if (result.compressedRanges > 0) deps.session.stats.lastInputTokens = currentTokens;
+  if (result.compressedRanges > 0) {
+    const textBaseline = result.payloadEstimate - (deps.imageFloor ?? 0);
+    if (textBaseline > deps.session.stats.lastInputTokens) {
+      deps.session.stats.lastInputTokens = textBaseline;
+      deps.session.stats.lastInputTokensSource = "estimate";
+    }
+  }
   result.savedTokens = Math.max(0, startTokens - currentTokens);
   if (currentTokens >= limit) result.failure = failure;
   result.fitsWindow = baselineKnown ? result.payloadEstimate < limit : finalUpper < limit;
@@ -54074,7 +54123,7 @@ var ROOT_CERT_FILE = "root-ca.pem";
 var ROOT_KEY_FILE = "root-ca-key.pem";
 var COMBINED_CA_FILE = "combined-ca.pem";
 var ROOT_CN = "billion-context MITM Root CA";
-var PLATFORM_CA_CANDIDATES = process.platform === "darwin" ? ["/etc/ssl/cert.pem", "/private/etc/ssl/cert.pem"] : [
+var PLATFORM_CA_CANDIDATES = process.platform === "darwin" ? ["/etc/ssl/cert.pem", "/private/etc/ssl/cert.pem"] : process.platform === "win32" ? [] : [
   "/etc/ssl/certs/ca-certificates.crt",
   "/etc/pki/tls/certs/ca-bundle.crt",
   "/etc/pki/ca-trust/extracted/pem/tls-ca-bundle.pem",
@@ -54870,6 +54919,7 @@ function resolveLearnedLimit(session, model) {
 function retractStaleLearnedLimits(session, model) {
   const x = session.stats?.lastInputTokens ?? 0;
   if (!(x > 0)) return false;
+  if (session.stats?.lastInputTokensSource !== "usage") return false;
   const stale = (v2) => typeof v2 === "number" && v2 > 0 && x - v2 >= Math.max(RETRACT_MIN_DELTA, v2 * RETRACT_MARGIN_PCT);
   const md = session.metadata ?? {};
   const removed = [];
@@ -54946,8 +54996,11 @@ function noteWeakOverflow(session, opts) {
       log("warn", `[${session.id}] weak overflow confirmed (${MIN_EVENTS}\xD7 high-usage failures, ${opts.reason}) \u2014 conservative window ${input} not below learned ${prev}; arming emergency shrink only`);
     }
   }
-  if (!session.stats) session.stats = { lastInputTokens: armInput };
-  else session.stats.lastInputTokens = Math.max(session.stats.lastInputTokens, armInput);
+  if (!session.stats) session.stats = { lastInputTokens: armInput, lastInputTokensSource: "estimate" };
+  else if (armInput > session.stats.lastInputTokens) {
+    session.stats.lastInputTokens = armInput;
+    session.stats.lastInputTokensSource = "estimate";
+  }
   markDirty(session);
 }
 
@@ -54988,6 +55041,7 @@ function warnCacheCollapse(session, input, cached) {
 
 // src/error-dump.ts
 import fs4 from "fs";
+import path11 from "path";
 var DEFAULT_MAX_BYTES = 2 * 1024 * 1024;
 var failCount = 0;
 var lastFailLog = 0;
@@ -55020,10 +55074,10 @@ function dumpRejectedBody(status, sessionId, body) {
         text = raw;
       }
     }
-    const dir = process.env.ACP_DUMP_DIR || `${stateDir()}/dumps`;
+    const dir = dumpsDir();
     fs4.mkdirSync(dir, { recursive: true });
     const sid = sessionId.replace(/[^a-zA-Z0-9_-]/g, "_");
-    const out = `${dir}/err-${Date.now()}-${sid}-${status}.json`;
+    const out = path11.join(dir, `err-${Date.now()}-${sid}-${status}.json`);
     fs4.writeFileSync(out, `${text}${marker}`);
     log("info", `[dump] upstream ${status} rejected body written to ${out}`);
     return out;
@@ -55225,6 +55279,7 @@ function recordUsage(ctx, usage, round) {
   if (total > 0) ctx.session.stats.inputTokens += total;
   if (total > 0) {
     ctx.session.stats.lastInputTokens = Math.max(0, total - (ctx.session.stats.compressCreditTokens ?? 0));
+    ctx.session.stats.lastInputTokensSource = "usage";
   }
   if (typeof cached === "number" && total > 0) {
     ctx.session.stats.cachedTokens += cached;
@@ -55596,10 +55651,11 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
       if (process.env.ACP_DUMP_BODY === "1") {
         try {
           const fs14 = await import("fs");
-          const dumpDir = process.env.ACP_DUMP_DIR || `${process.env.HOME}/.local/state/billion-context/dumps`;
+          const path22 = await import("path");
+          const dumpDir = dumpsDir();
           fs14.mkdirSync(dumpDir, { recursive: true });
-          const sid = ctx.session.id ?? "unknown";
-          fs14.writeFileSync(`${dumpDir}/req-${Date.now()}-${sid}-REREQUEST.json`, JSON.stringify(newBody, null, 2));
+          const sid = (ctx.session.id ?? "unknown").replace(/[^a-zA-Z0-9_-]/g, "_");
+          fs14.writeFileSync(path22.join(dumpDir, `req-${Date.now()}-${sid}-REREQUEST.json`), JSON.stringify(newBody, null, 2));
         } catch {
         }
       }
@@ -58431,10 +58487,10 @@ var prefixAffinity = new PrefixAffinityResolver();
 
 // src/affinity-persist.ts
 import fs5 from "fs";
-import path11 from "path";
+import path12 from "path";
 var PERSIST_DEBOUNCE_MS = 5e3;
 function affinityFile() {
-  return path11.join(stateDir(), "prefix-affinity.json");
+  return path12.join(stateDir(), "prefix-affinity.json");
 }
 var timer = null;
 var writing = false;
@@ -58445,7 +58501,7 @@ function writeSnapshot() {
     const file = affinityFile();
     const snapshot = { version: 1, entries: prefixAffinity.exportSnapshot() };
     const tmp = `${file}.tmp`;
-    fs5.mkdirSync(path11.dirname(file), { recursive: true });
+    fs5.mkdirSync(path12.dirname(file), { recursive: true });
     fs5.writeFileSync(tmp, JSON.stringify(snapshot));
     fs5.renameSync(tmp, file);
   } catch (e) {
@@ -58475,7 +58531,7 @@ function hydratePrefixAffinity() {
     if (!fs5.existsSync(file)) return;
     const parsed = JSON.parse(fs5.readFileSync(file, "utf8"));
     const imported = prefixAffinity.importSnapshot(parsed.entries);
-    if (imported > 0) log("info", `[prefix-affinity] hydrated ${imported} chain(s) from ${path11.basename(file)} \u2014 anonymous sessions reattach across restarts`);
+    if (imported > 0) log("info", `[prefix-affinity] hydrated ${imported} chain(s) from ${path12.basename(file)} \u2014 anonymous sessions reattach across restarts`);
   } catch (e) {
     log("warn", `[prefix-affinity] hydrate failed (${e instanceof Error ? e.message : String(e)}); starting with empty affinity`);
   }
@@ -58625,11 +58681,11 @@ function buildStatusPanel(input) {
 // src/plugin.ts
 import { fileURLToPath as fileURLToPath3 } from "url";
 import fs6 from "fs";
-import path12 from "path";
+import path13 from "path";
 var PROXY_VERSION = (() => {
   try {
     const here = fileURLToPath3(import.meta.url);
-    const pkg = path12.join(path12.dirname(here), "..", "package.json");
+    const pkg = path13.join(path13.dirname(here), "..", "package.json");
     return JSON.parse(fs6.readFileSync(pkg, "utf8")).version ?? "dev";
   } catch {
     return "dev";
@@ -58642,7 +58698,7 @@ var PLUGIN_PROTOCOL_VERSION = 1;
 var VERSION2 = (() => {
   try {
     const here = fileURLToPath3(import.meta.url);
-    const pkg = path12.join(path12.dirname(here), "..", "package.json");
+    const pkg = path13.join(path13.dirname(here), "..", "package.json");
     return JSON.parse(fs6.readFileSync(pkg, "utf8")).version ?? "dev";
   } catch {
     return "dev";
@@ -58672,7 +58728,7 @@ function pluginReportedContextWindow(headers) {
 var MAX_PLUGIN_CONVERSATIONS = 1024;
 var conversations = /* @__PURE__ */ new Map();
 var remembered = /* @__PURE__ */ new Map();
-var conversationsFile = () => path12.join(stateDir(), "plugin-conversations.json");
+var conversationsFile = () => path13.join(stateDir(), "plugin-conversations.json");
 var conversationsSaveTimer;
 var conversationsDirty = false;
 function writeConversationsFile() {
@@ -58924,7 +58980,7 @@ function handlePluginStatus(conversationId2, res, deps, fallbackLatest = false)
       nudge = deps.core.processTurn({
         messages,
         state: session.state,
-        config: deps.config,
+        config: effectiveConfig(session, deps.config),
         tokenCount: session.stats.lastInputTokens,
         renderTags: "none"
       }).nudge;
@@ -59012,7 +59068,9 @@ async function handlePluginTool(payload, res, deps) {
       const messages = mem ? mem.processed.length > 0 ? mem.processed : mem.original : [];
       return executeProxyTool(tool, args, {
         core: deps.core,
-        config: deps.config,
+        // #833: run proxy tools under the session's last resolved Config
+        // (same values the wire path used), not the base kernelConfig.
+        config: effectiveConfig(session, deps.config),
         messages,
         session,
         log: (m2) => deps.log("info", `[${session.id}] [plugin] ${m2}`)
@@ -59098,6 +59156,7 @@ function applyUsageSample(session, sample, protocol) {
   if (sample.inputTokens !== void 0 && total > 0) {
     session.stats.inputTokens += total;
     session.stats.lastInputTokens = Math.max(0, total - (session.stats.compressCreditTokens ?? 0));
+    session.stats.lastInputTokensSource = "usage";
     warnCacheCollapse(session, total, sample.cachedTokens ?? 0);
     const hit = sample.cachedTokens === void 0 ? void 0 : Math.round(100 * (sample.cachedTokens ?? 0) / total);
     const foldNew = session.stats.pendingFoldUsage === true;
@@ -59620,12 +59679,12 @@ import tls3 from "tls";
 // src/discover.ts
 import fs8 from "fs";
 import os2 from "os";
-import path14 from "path";
+import path15 from "path";
 
 // src/client-config.ts
 import fs7 from "fs";
 import os from "os";
-import path13 from "path";
+import path14 from "path";
 
 // node_modules/jsonc-parser/lib/esm/impl/scanner.js
 function createScanner(text, ignoreTrivia = false) {
@@ -60502,8 +60561,8 @@ function qoderIsCnSite(env = process.env) {
   if (site === "cn") return true;
   if (nonEmpty2(env.QODERCN_CONFIG_DIR) || nonEmpty2(env.QODERCN_CLI_HOME)) return true;
   const h = os.homedir();
-  const cnDir = path13.join(h, ".qoder-cn");
-  const intlDir = path13.join(h, ".qoder");
+  const cnDir = path14.join(h, ".qoder-cn");
+  const intlDir = path14.join(h, ".qoder");
   try {
     if (fs7.existsSync(cnDir) && !fs7.existsSync(intlDir)) return true;
   } catch {
@@ -60519,11 +60578,11 @@ function resolveQoderHome(env = process.env) {
   const cliHome = nonEmpty2(cliHomeEnv) ? cliHomeEnv : h;
   const dirNameEnv = cn2 ? env.QODERCN_CONFIG_DIR_NAME : env.QODER_CONFIG_DIR_NAME;
   const dirName = nonEmpty2(dirNameEnv) ? dirNameEnv : cn2 ? ".qoder-cn" : ".qoder";
-  return path13.join(cliHome, dirName);
+  return path14.join(cliHome, dirName);
 }
 function readQoderConfig(qoderHome, env = process.env) {
   const result = {};
-  const obj = readJsonObject(path13.join(qoderHome, "settings.json"));
+  const obj = readJsonObject(path14.join(qoderHome, "settings.json"));
   const model = obj?.model;
   if (typeof model === "string" && model.trim().length > 0) {
     result.model = model.trim();
@@ -60553,27 +60612,27 @@ function readJsonObject(filePath) {
 }
 function resolvePiHome(env) {
   const h = os.homedir();
-  return nonEmpty2(env.PI_CODING_AGENT_DIR) ? env.PI_CODING_AGENT_DIR : nonEmpty2(env.PI_HOME) ? env.PI_HOME : path13.join(h, ".pi", "agent");
+  return nonEmpty2(env.PI_CODING_AGENT_DIR) ? env.PI_CODING_AGENT_DIR : nonEmpty2(env.PI_HOME) ? env.PI_HOME : path14.join(h, ".pi", "agent");
 }
 function resolveOmpHome(env) {
   const h = os.homedir();
-  return nonEmpty2(env.PI_CODING_AGENT_DIR) ? env.PI_CODING_AGENT_DIR : path13.join(h, ".omp", "agent");
+  return nonEmpty2(env.PI_CODING_AGENT_DIR) ? env.PI_CODING_AGENT_DIR : path14.join(h, ".omp", "agent");
 }
 function resolveHermesHome(env) {
   const h = os.homedir();
-  return nonEmpty2(env.HERMES_HOME) ? env.HERMES_HOME : path13.join(h, ".hermes");
+  return nonEmpty2(env.HERMES_HOME) ? env.HERMES_HOME : path14.join(h, ".hermes");
 }
 function resolveDshHome(env) {
   const h = os.homedir();
-  return nonEmpty2(env.DSH_HOME) ? env.DSH_HOME : path13.join(h, ".dsh");
+  return nonEmpty2(env.DSH_HOME) ? env.DSH_HOME : path14.join(h, ".dsh");
 }
 function resolveCodexHome(env) {
   const h = os.homedir();
-  return nonEmpty2(env.CODEX_HOME) ? env.CODEX_HOME : path13.join(h, ".codex");
+  return nonEmpty2(env.CODEX_HOME) ? env.CODEX_HOME : path14.join(h, ".codex");
 }
 function resolveCodebuddyHome(env) {
   const h = os.homedir();
-  return nonEmpty2(env.CODEBUDDY_CONFIG_DIR) ? env.CODEBUDDY_CONFIG_DIR : path13.join(h, ".codebuddy");
+  return nonEmpty2(env.CODEBUDDY_CONFIG_DIR) ? env.CODEBUDDY_CONFIG_DIR : path14.join(h, ".codebuddy");
 }
 function parseCodebuddyModelsJson(obj) {
   const out = { models: [], urls: [] };
@@ -60627,7 +60686,7 @@ function readCodebuddyConfig(codebuddyHome, cwd, env = process.env) {
   let codebuddyBaseUrl;
   let model;
   let autoCompactWindow;
-  const settings = readJsonObject(path13.join(codebuddyHome, "settings.json"));
+  const settings = readJsonObject(path14.join(codebuddyHome, "settings.json"));
   const settingsEnv = settings?.env;
   if (settingsEnv && typeof settingsEnv === "object" && !Array.isArray(settingsEnv)) {
     const e = settingsEnv;
@@ -60643,8 +60702,8 @@ function readCodebuddyConfig(codebuddyHome, cwd, env = process.env) {
   const urls = [];
   const seenUrl = /* @__PURE__ */ new Set();
   for (const f2 of [
-    path13.join(codebuddyHome, "models.json"),
-    path13.join(cwd, ".codebuddy", "models.json")
+    path14.join(codebuddyHome, "models.json"),
+    path14.join(cwd, ".codebuddy", "models.json")
   ]) {
     const parsed = parseCodebuddyModelsJson(readJsonFile(f2));
     for (const w2 of parsed.models) windowByModel.set(w2.id, w2.contextWindow);
@@ -60680,7 +60739,7 @@ function parseDshSettingsYaml(text) {
 function readDshConfig(dshHome) {
   let text;
   try {
-    text = fs7.readFileSync(path13.join(dshHome, "settings.yaml"), "utf8");
+    text = fs7.readFileSync(path14.join(dshHome, "settings.yaml"), "utf8");
   } catch {
     return { baseUrls: [] };
   }
@@ -60695,7 +60754,7 @@ var JCODE_DEFAULT_MODEL_HOSTS = [
 ];
 function resolveTraeHome(env) {
   const h = os.homedir();
-  return nonEmpty2(env.TRAE_CONFIG_DIR) ? env.TRAE_CONFIG_DIR : path13.join(h, ".trae");
+  return nonEmpty2(env.TRAE_CONFIG_DIR) ? env.TRAE_CONFIG_DIR : path14.join(h, ".trae");
 }
 function readTraeConfig(env) {
   const result = {};
@@ -60705,8 +60764,8 @@ function readTraeConfig(env) {
 }
 function readClaudeSettings(homeDir, cwd, env = process.env) {
   const files = [
-    path13.join(homeDir, ".claude", "settings.json"),
-    path13.join(cwd, ".claude", "settings.json")
+    path14.join(homeDir, ".claude", "settings.json"),
+    path14.join(cwd, ".claude", "settings.json")
   ];
   let anthropicBaseUrl;
   let model;
@@ -60779,7 +60838,7 @@ function parseCodexToml(text) {
   return result;
 }
 function readCodexConfig(codexHome) {
-  const cfgPath = path13.join(codexHome, "config.toml");
+  const cfgPath = path14.join(codexHome, "config.toml");
   let text;
   try {
     text = fs7.readFileSync(cfgPath, "utf8");
@@ -60860,10 +60919,10 @@ function parseKimiToml(text) {
   return result;
 }
 function resolveKimiHome(env = process.env) {
-  return nonEmpty2(env.KIMI_CODE_HOME) ? env.KIMI_CODE_HOME : path13.join(os.homedir(), ".kimi-code");
+  return nonEmpty2(env.KIMI_CODE_HOME) ? env.KIMI_CODE_HOME : path14.join(os.homedir(), ".kimi-code");
 }
 function readKimiConfig(kimiHome, env = process.env) {
-  const cfgPath = path13.join(kimiHome, "config.toml");
+  const cfgPath = path14.join(kimiHome, "config.toml");
   let text;
   try {
     text = fs7.readFileSync(cfgPath, "utf8");
@@ -60883,7 +60942,7 @@ function readKimiConfig(kimiHome, env = process.env) {
   return config;
 }
 function readPiConfig(piHome) {
-  const cfgPath = path13.join(piHome, "models.json");
+  const cfgPath = path14.join(piHome, "models.json");
   const obj = readJsonObject(cfgPath);
   const providers = {};
   const rawProviders = obj?.providers;
@@ -60964,7 +61023,7 @@ function parseOmpYaml(text) {
   return result;
 }
 function readOmpConfig(ompHome) {
-  const cfgPath = path13.join(ompHome, "models.yml");
+  const cfgPath = path14.join(ompHome, "models.yml");
   let text;
   try {
     text = fs7.readFileSync(cfgPath, "utf8");
@@ -61045,7 +61104,7 @@ function parseHermesYaml(text) {
   return result;
 }
 function readHermesConfig(hermesHome) {
-  const cfgPath = path13.join(hermesHome, "config.yaml");
+  const cfgPath = path14.join(hermesHome, "config.yaml");
   let text;
   try {
     text = fs7.readFileSync(cfgPath, "utf8");
@@ -61082,10 +61141,10 @@ function mergeConfigDeep(target, source) {
 }
 function readOpencodeConfigRoot(env) {
   let root;
-  const xdg2 = nonEmpty2(env.XDG_CONFIG_HOME) ? env.XDG_CONFIG_HOME : path13.join(os.homedir(), ".config");
-  const dir = path13.join(xdg2, "opencode");
+  const xdg2 = nonEmpty2(env.XDG_CONFIG_HOME) ? env.XDG_CONFIG_HOME : path14.join(os.homedir(), ".config");
+  const dir = path14.join(xdg2, "opencode");
   for (const file of ["config.json", "opencode.json", "opencode.jsonc"]) {
-    const next = readConfigFileRoot(path13.join(dir, file));
+    const next = readConfigFileRoot(path14.join(dir, file));
     if (next !== void 0) root = root === void 0 ? next : mergeConfigDeep(root, next);
   }
   if (nonEmpty2(env.OPENCODE_CONFIG)) {
@@ -61098,10 +61157,10 @@ function opencodePluginBaseDir(env, key) {
   const oc = env.OPENCODE_CONFIG;
   if (nonEmpty2(oc)) {
     const parsed = readConfigFileRoot(oc);
-    if (parsed !== void 0 && key in parsed) return path13.dirname(oc);
+    if (parsed !== void 0 && key in parsed) return path14.dirname(oc);
   }
-  const xdg2 = nonEmpty2(env.XDG_CONFIG_HOME) ? env.XDG_CONFIG_HOME : path13.join(os.homedir(), ".config");
-  return path13.join(xdg2, "opencode");
+  const xdg2 = nonEmpty2(env.XDG_CONFIG_HOME) ? env.XDG_CONFIG_HOME : path14.join(os.homedir(), ".config");
+  return path14.join(xdg2, "opencode");
 }
 function parseOpencodeProviders(parsed) {
   const providers = {};
@@ -61151,7 +61210,7 @@ function parseZcodeConfig(obj) {
   return result;
 }
 function readZcodeConfig(zcodeHome) {
-  const cfgPath = path13.join(zcodeHome, "v2", "config.json");
+  const cfgPath = path14.join(zcodeHome, "v2", "config.json");
   let txt;
   try {
     txt = fs7.readFileSync(cfgPath, "utf8");
@@ -61170,10 +61229,10 @@ function loadClientConfig(env, cwd) {
   const home = os.homedir();
   const config = {};
   config.claude = readClaudeSettings(home, cwd, env);
-  const codexHome = nonEmpty2(env.CODEX_HOME) ? env.CODEX_HOME : path13.join(home, ".codex");
+  const codexHome = nonEmpty2(env.CODEX_HOME) ? env.CODEX_HOME : path14.join(home, ".codex");
   config.codex = readCodexConfig(codexHome);
   config.pi = readPiConfig(resolvePiHome(env));
-  const zcodeHome = nonEmpty2(env.ZCODE_DATA_BASE_DIR) ? env.ZCODE_DATA_BASE_DIR : path13.join(home, ".zcode");
+  const zcodeHome = nonEmpty2(env.ZCODE_DATA_BASE_DIR) ? env.ZCODE_DATA_BASE_DIR : path14.join(home, ".zcode");
   config.zcode = readZcodeConfig(zcodeHome);
   config.omp = readOmpConfig(resolveOmpHome(env));
   config.opencode = parseOpencodeProviders(readOpencodeConfigRoot(env));
@@ -61260,20 +61319,20 @@ function extractHttpsHosts(config) {
 }
 function configFilePaths(env) {
   const home = os2.homedir();
-  const codexHome = nonEmpty2(env.CODEX_HOME) ? env.CODEX_HOME : path14.join(home, ".codex");
-  const zcodeHome = nonEmpty2(env.ZCODE_DATA_BASE_DIR) ? env.ZCODE_DATA_BASE_DIR : path14.join(home, ".zcode");
+  const codexHome = nonEmpty2(env.CODEX_HOME) ? env.CODEX_HOME : path15.join(home, ".codex");
+  const zcodeHome = nonEmpty2(env.ZCODE_DATA_BASE_DIR) ? env.ZCODE_DATA_BASE_DIR : path15.join(home, ".zcode");
   const codebuddyHome = resolveCodebuddyHome(env);
   return [
-    path14.join(home, ".claude", "settings.json"),
-    path14.join(process.cwd(), ".claude", "settings.json"),
-    path14.join(codexHome, "config.toml"),
-    path14.join(resolvePiHome(env), "models.json"),
-    path14.join(zcodeHome, "v2", "config.json"),
-    path14.join(codebuddyHome, "settings.json"),
-    path14.join(codebuddyHome, "models.json"),
-    path14.join(process.cwd(), ".codebuddy", "models.json"),
-    path14.join(resolveQoderHome(env), "settings.json"),
-    path14.join(resolveTraeHome(env), "traecli.yaml")
+    path15.join(home, ".claude", "settings.json"),
+    path15.join(process.cwd(), ".claude", "settings.json"),
+    path15.join(codexHome, "config.toml"),
+    path15.join(resolvePiHome(env), "models.json"),
+    path15.join(zcodeHome, "v2", "config.json"),
+    path15.join(codebuddyHome, "settings.json"),
+    path15.join(codebuddyHome, "models.json"),
+    path15.join(process.cwd(), ".codebuddy", "models.json"),
+    path15.join(resolveQoderHome(env), "settings.json"),
+    path15.join(resolveTraeHome(env), "traecli.yaml")
   ];
 }
 function readMtimes(paths) {
@@ -62324,8 +62383,8 @@ function logUnrecognizedPath(log2, url) {
     log2("info", `unrecognized path ${key}: forwarding unchanged; further occurrences suppressed`);
   }
 }
-function isModelDiscoveryPath(path20) {
-  return path20.replace(/\/+$/, "").endsWith("/models");
+function isModelDiscoveryPath(path22) {
+  return path22.replace(/\/+$/, "").endsWith("/models");
 }
 
 // src/server/context-window.ts
@@ -63109,7 +63168,7 @@ async function handle(req, res, opts, core, config, log2, instanceId, instanceSt
   }
   if (bodyDumpEnabled() && parsed && typeof parsed === "object") {
     try {
-      const rawDir = process.env.ACP_RAW_DUMP_DIR || `${stateDir()}/raw`;
+      const rawDir = process.env.ACP_RAW_DUMP_DIR || path16.join(stateDir(), "raw");
       try {
         fs9.mkdirSync(rawDir, { recursive: true });
       } catch {
@@ -63118,7 +63177,7 @@ async function handle(req, res, opts, core, config, log2, instanceId, instanceSt
         Object.fromEntries(Object.entries(req.headers).map(([k2, v2]) => [k2, Array.isArray(v2) ? v2.join(",") : String(v2)]))
       );
       const hdrText = Object.entries(hdrs).map(([k2, v2]) => `${k2}: ${v2}`).join("\n");
-      fs9.writeFileSync(`${rawDir}/${Date.now()}-INCOMING.txt`, `${req.method} ${maskUrlsInText(req.url ?? "")}
+      fs9.writeFileSync(path16.join(rawDir, `${Date.now()}-INCOMING.txt`), `${req.method} ${maskUrlsInText(req.url ?? "")}
 ${hdrText}
 
 ${bodyBuffer.toString("utf8")}`);
@@ -63323,7 +63382,7 @@ ${bodyBuffer.toString("utf8")}`);
       const prevInput = session.stats.lastInputTokens ?? 0;
       const prevWindow = session.metadata.lastTurnWindow;
       const resolved = reqConfig.modelContextLimit;
-      if (prevWindow !== void 0 && prevInput > prevWindow && prevInput > resolved && prevInput >= 1e3) {
+      if (session.stats.lastInputTokensSource === "usage" && prevWindow !== void 0 && prevInput > prevWindow && prevInput > resolved && prevInput >= 1e3) {
         const map = session.metadata.confirmedContextLimits ?? {};
         const prev = map[reqModel];
         if (prev === void 0 || prevInput > prev) {
@@ -63350,6 +63409,7 @@ ${bodyBuffer.toString("utf8")}`);
     }
     session.metadata.effectiveContextLimit = reqConfig.modelContextLimit;
     session.metadata.lastTurnWindow = reqConfig.modelContextLimit;
+    storeEffectiveConfig(session, reqConfig);
     acquireInFlight(session);
     try {
       await withSessionLock(session, async () => {
@@ -63393,6 +63453,7 @@ ${bodyBuffer.toString("utf8")}`);
             opts,
             core,
             reqConfig,
+            nativeWindow,
             parsed.model,
             route,
             affinity,
@@ -64088,8 +64149,8 @@ function logUpstreamProxyDecision(opts, upstreamUrl, decision) {
   const via = decision.proxy ? `via ${maskUrlForLog(decision.proxy)}` : "direct";
   logMsg(opts, "info", `[upstream-proxy] ${maskHostPortForLog(host)} ${via} (source=${decision.source})`);
 }
-function inferWireProtocol(path20) {
-  const p2 = path20.split("?", 2)[0];
+function inferWireProtocol(path22) {
+  const p2 = path22.split("?", 2)[0];
   if (p2.endsWith("/chat/completions") || p2.endsWith("/llm_raw_chat")) return "openai";
   if (p2.endsWith("/responses") || p2.endsWith("/responses/compact")) return "responses";
   return null;
@@ -64172,7 +64233,7 @@ function beginPreflightHold(res, prepared, log2) {
   }, PREFLIGHT_KEEPALIVE_MS);
   return () => clearInterval(iv);
 }
-async function preflightCompressIfNeeded(prepared, runPrepare, req, res, opts, core, config, model, route, affinity, anonymous, log2, instanceId) {
+async function preflightCompressIfNeeded(prepared, runPrepare, req, res, opts, core, config, configuredWindow, model, route, affinity, anonymous, log2, instanceId) {
   const session = prepared.session;
   const limit = config.modelContextLimit;
   const compressionTarget = prepared.protocol === "responses" && isCodexClient(req.headers) && codexCompactMode() === "intercept" ? limit * CODEX_COMPACT_HEALTH_RATIO : limit;
@@ -64187,7 +64248,8 @@ async function preflightCompressIfNeeded(prepared, runPrepare, req, res, opts, c
   if (limit <= 0 || !model || tokenCount < compressionTarget) return prepared;
   const payloadFitsWindow = (unknownBaseline ? tokenCount : payloadEstimate) < limit;
   const learnedLimit = resolveLearnedLimit(session, model);
-  const noOverflowEvidence = session.stats.lastInputTokens < limit && learnedLimit === void 0;
+  const governingLearned = learnedLimit !== void 0 && learnedLimit <= configuredWindow ? learnedLimit : void 0;
+  const noOverflowEvidence = (session.stats.lastInputTokens < limit || session.stats.lastInputTokensSource !== "usage") && governingLearned === void 0;
   if (imageTokens > 0 && payloadEstimate >= limit && textEstimate < limit && noOverflowEvidence) {
     log2("warn", `[${session.id}] image-dominated payload (~${textEstimate} text + ~${imageTokens} image tokens) exceeds window ${limit} by estimate only, no upstream overflow evidence \u2014 forwarding once so the upstream arbitrates billing (#496)`);
     return prepared;
@@ -64301,6 +64363,7 @@ function armFailureShrink(prepared, log2, reason) {
   if (!Number.isFinite(est) || est <= 0) return;
   if (est > s3.stats.lastInputTokens) {
     s3.stats.lastInputTokens = est;
+    s3.stats.lastInputTokensSource = "estimate";
     markDirty(s3);
     log2("warn", `[${s3.id}] ${reason} with no usage report \u2014 armed emergency shrink with local estimate ${est} tokens`);
   }
@@ -64362,13 +64425,13 @@ async function forward(req, res, opts, body, prepared, core, config, log2, route
         log2("info", `[debug] tools=[${toolNames.join(",")}] msgs=${parsed.messages?.length ?? 0} stream=${parsed.stream ?? false} system_len=${JSON.stringify(parsed.messages?.find((m2) => m2.role === "system")?.content ?? "").length}`);
       }
       if (bodyDumpEnabled() && process.env.ACP_DUMP_REQ !== "0") {
-        const dumpDir = process.env.ACP_DUMP_DIR || `${stateDir()}/dumps`;
+        const dumpDir = dumpsDir();
         try {
           fs9.mkdirSync(dumpDir, { recursive: true });
         } catch {
         }
         const sid = prepared?.session.id ?? "unknown";
-        const out = `${dumpDir}/req-${Date.now()}-${safeSessionId(sid)}.json`;
+        const out = path16.join(dumpDir, `req-${Date.now()}-${safeSessionId(sid)}.json`);
         try {
           const pretty = JSON.stringify(JSON.parse(wireBody), null, 2);
           fs9.writeFileSync(out, pretty);
@@ -64392,9 +64455,9 @@ async function forward(req, res, opts, body, prepared, core, config, log2, route
   }
   const rawBase = bodyDumpEnabled() ? (() => {
     try {
-      const rawDir = process.env.ACP_RAW_DUMP_DIR || `${stateDir()}/raw`;
+      const rawDir = process.env.ACP_RAW_DUMP_DIR || path16.join(stateDir(), "raw");
       fs9.mkdirSync(rawDir, { recursive: true });
-      return `${rawDir}/${Date.now()}-${safeSessionId(prepared?.session.id)}`;
+      return path16.join(rawDir, `${Date.now()}-${safeSessionId(prepared?.session.id)}`);
     } catch {
       return "";
     }
@@ -64570,9 +64633,13 @@ ${hdrText}
         }
         if (info.window) {
           s3.stats.lastInputTokens = info.window;
+          s3.stats.lastInputTokensSource = "usage";
         } else {
           const floor = (reqModel ? confirmedMap[reqModel] : void 0) ?? s3.metadata.confirmedContextLimit ?? s3.metadata.effectiveContextLimit ?? 0;
-          if (floor > 0) s3.stats.lastInputTokens = Math.max(s3.stats.lastInputTokens, floor);
+          if (floor > 0 && floor > s3.stats.lastInputTokens) {
+            s3.stats.lastInputTokens = floor;
+            s3.stats.lastInputTokensSource = "usage";
+          }
         }
         markDirty(s3);
       }
@@ -64838,6 +64905,7 @@ ${buildAbsorbSystemPrompt(absorbToolName(loopConfig))}` : "";
               0,
               total - (prepared.session.stats.compressCreditTokens ?? 0)
             );
+            prepared.session.stats.lastInputTokensSource = "usage";
             if (typeof cached === "number") {
               prepared.session.stats.cachedTokens += cached;
               prepared.session.stats.cacheSamples += 1;
@@ -67969,7 +68037,7 @@ var To = (s3) => {
 };
 
 // src/update.ts
-import path15 from "path";
+import path17 from "path";
 import { fileURLToPath as fileURLToPath4 } from "url";
 var REGISTRY_BASE = "https://registry.npmjs.org";
 function normalizeUpdateTag(tag) {
@@ -67979,8 +68047,8 @@ function registryUrlFor(packageName, tag) {
   return `${REGISTRY_BASE}/${packageName}/${encodeURIComponent(tag)}`;
 }
 var CHECK_INTERVAL_MS = 3 * 60 * 1e3;
-var THROTTLE_FILE = path15.join(cacheDir(), ".update-check");
-var LOCK_FILE = path15.join(cacheDir(), ".update-lock");
+var THROTTLE_FILE = path17.join(cacheDir(), ".update-check");
+var LOCK_FILE = path17.join(cacheDir(), ".update-lock");
 var LOCK_MAX_AGE_MS = 30 * 60 * 1e3;
 function shouldStealLock(holderAlive, ageMs) {
   return !holderAlive || ageMs >= LOCK_MAX_AGE_MS;
@@ -68051,27 +68119,27 @@ async function readLastCheck() {
 }
 async function writeLastCheck(ts2) {
   try {
-    await mkdir2(path15.dirname(THROTTLE_FILE), { recursive: true });
+    await mkdir2(path17.dirname(THROTTLE_FILE), { recursive: true });
     await writeFile2(THROTTLE_FILE, String(ts2), "utf-8");
   } catch {
   }
 }
 async function findInstallDir(packageName) {
-  let dir = path15.dirname(fileURLToPath4(import.meta.url));
+  let dir = path17.dirname(fileURLToPath4(import.meta.url));
   for (; ; ) {
     try {
-      const pkg = JSON.parse(await readFile3(path15.join(dir, "package.json"), "utf-8"));
+      const pkg = JSON.parse(await readFile3(path17.join(dir, "package.json"), "utf-8"));
       if (pkg.name === packageName) return dir;
     } catch {
     }
-    const parent = path15.dirname(dir);
+    const parent = path17.dirname(dir);
     if (parent === dir) return void 0;
     dir = parent;
   }
 }
 async function isGitWorkingTree(dir) {
   try {
-    await access(path15.join(dir, ".git"));
+    await access(path17.join(dir, ".git"));
     return true;
   } catch {
     return false;
@@ -68079,7 +68147,7 @@ async function isGitWorkingTree(dir) {
 }
 async function readDiskVersion(installDir) {
   try {
-    const pkg = JSON.parse(await readFile3(path15.join(installDir, "package.json"), "utf-8"));
+    const pkg = JSON.parse(await readFile3(path17.join(installDir, "package.json"), "utf-8"));
     return pkg.version;
   } catch {
     return void 0;
@@ -68116,13 +68184,13 @@ async function syntaxCheckEntry(entryAbs) {
   } catch (e) {
     return `entry unreadable: ${String(e)}`;
   }
-  const tmpCheck = path15.join(cacheDir(), ".update-syntax-check.mjs");
+  const tmpCheck = path17.join(cacheDir(), ".update-syntax-check.mjs");
   try {
     await mkdir2(cacheDir(), { recursive: true });
     await writeFile2(tmpCheck, source);
     const r = await runNodeCheck(tmpCheck);
     if (r.code !== 0) {
-      return `entry does not parse (${path15.basename(entryAbs)}): ${r.stderr.split("\n").filter(Boolean).slice(0, 3).join(" | ").slice(0, 300)}`;
+      return `entry does not parse (${path17.basename(entryAbs)}): ${r.stderr.split("\n").filter(Boolean).slice(0, 3).join(" | ").slice(0, 300)}`;
     }
     return null;
   } finally {
@@ -68135,7 +68203,7 @@ async function syntaxCheckEntry(entryAbs) {
 async function verifyEntries(dir, label) {
   let pkg;
   try {
-    pkg = JSON.parse(await readFile3(path15.join(dir, "package.json"), "utf-8"));
+    pkg = JSON.parse(await readFile3(path17.join(dir, "package.json"), "utf-8"));
   } catch (e) {
     return `${label}: package.json unreadable: ${String(e)}`;
   }
@@ -68145,11 +68213,11 @@ async function verifyEntries(dir, label) {
   }
   for (const rel2 of entries) {
     try {
-      await access(path15.join(dir, rel2));
+      await access(path17.join(dir, rel2));
     } catch {
       return `${label}: entry missing: ${rel2}`;
     }
-    const reason = await syntaxCheckEntry(path15.join(dir, rel2));
+    const reason = await syntaxCheckEntry(path17.join(dir, rel2));
     if (reason) return `${label}: ${reason}`;
   }
   return null;
@@ -68366,14 +68434,14 @@ async function installViaTarball(version2, tarballUrl, installDir, integrity, sh
   if (!v2.ok) {
     return { ok: false, error: `tarball integrity verification failed: ${v2.error}` };
   }
-  const tmpFile = path15.join(cacheDir(), `.update-${version2}.tgz`);
+  const tmpFile = path17.join(cacheDir(), `.update-${version2}.tgz`);
   try {
     await mkdir2(cacheDir(), { recursive: true });
     await writeFile2(tmpFile, tgzBuffer);
   } catch (e) {
     return { ok: false, error: `failed to write temp file ${tmpFile}: ${String(e)}` };
   }
-  const stagingDir = path15.join(cacheDir(), `.update-staging-${version2}`);
+  const stagingDir = path17.join(cacheDir(), `.update-staging-${version2}`);
   try {
     await rm2(stagingDir, { recursive: true, force: true });
     await mkdir2(stagingDir, { recursive: true });
@@ -68395,7 +68463,7 @@ async function installViaTarball(version2, tarballUrl, installDir, integrity, sh
   } finally {
     await rm2(tmpFile, { force: true });
   }
-  const backupDir = path15.join(cacheDir(), `.update-backup-${version2}`);
+  const backupDir = path17.join(cacheDir(), `.update-backup-${version2}`);
   try {
     await rm2(backupDir, { recursive: true, force: true });
     await cp(installDir, backupDir, { recursive: true, force: true });
@@ -68452,12 +68520,12 @@ function startAutoUpdate(opts) {
 
 // src/mcp.ts
 import fs11 from "fs";
-import path16 from "path";
+import path18 from "path";
 import { fileURLToPath as fileURLToPath5 } from "url";
 var VERSION3 = (() => {
   try {
     const here = fileURLToPath5(import.meta.url);
-    const pkg = path16.join(path16.dirname(here), "..", "package.json");
+    const pkg = path18.join(path18.dirname(here), "..", "package.json");
     return JSON.parse(fs11.readFileSync(pkg, "utf8")).version ?? "dev";
   } catch {
     return "dev";
@@ -68669,7 +68737,7 @@ if (process.argv[1] && /(?:^|[\\/])mcp\.(?:ts|js)$/.test(process.argv[1])) {
 
 // src/plugin-install.ts
 import fs12 from "fs";
-import path17 from "path";
+import path19 from "path";
 import os4 from "os";
 import { execFileSync as execFileSync2 } from "child_process";
 import { fileURLToPath as fileURLToPath6 } from "url";
@@ -68688,12 +68756,12 @@ function proxyOriginForInstall() {
 var PLUGIN_AGENTS = ["pi", "omp", "claude", "codex", "opencode"];
 function selfPackageRoot() {
   const here = fileURLToPath6(import.meta.url);
-  return path17.resolve(path17.dirname(here), "..");
+  return path19.resolve(path19.dirname(here), "..");
 }
 function homeFile(rel2, envOverride) {
   const raw = (envOverride !== void 0 ? process.env[envOverride] : void 0)?.trim();
   const base = raw && raw.length > 0 ? raw : os4.homedir();
-  return path17.join(base, rel2);
+  return path19.join(base, rel2);
 }
 function backupOnce(file) {
   if (fs12.existsSync(file) && !fs12.existsSync(`${file}.bili-bak`)) {
@@ -68712,7 +68780,7 @@ function readJson(file) {
   try {
     parsed = JSON.parse(text);
   } catch (err2) {
-    throw new Error(`${file}: not valid JSON (${err2 instanceof Error ? err2.message : String(err2)}) \u2014 fix it or restore ${path17.basename(file)}.bili-bak first; refusing to overwrite`);
+    throw new Error(`${file}: not valid JSON (${err2 instanceof Error ? err2.message : String(err2)}) \u2014 fix it or restore ${path19.basename(file)}.bili-bak first; refusing to overwrite`);
   }
   if (typeof parsed !== "object" || parsed === null || Array.isArray(parsed)) {
     throw new Error(`${file}: expected a JSON object at top level, refusing to overwrite`);
@@ -68720,7 +68788,7 @@ function readJson(file) {
   return parsed;
 }
 function writeJson(file, data) {
-  fs12.mkdirSync(path17.dirname(file), { recursive: true });
+  fs12.mkdirSync(path19.dirname(file), { recursive: true });
   backupOnce(file);
   fs12.writeFileSync(file, JSON.stringify(data, null, 2) + "\n");
 }
@@ -68731,7 +68799,7 @@ function requireDistFile(file) {
   }
 }
 function piSettingsFile() {
-  return path17.join(resolvePiHome(process.env), "settings.json");
+  return path19.join(resolvePiHome(process.env), "settings.json");
 }
 function isPiEntry(entry, root) {
   return entry === root || /^npm:billion-context(-pi)?(@|$)/.test(entry) || /(^|[/\\])node_modules[/\\]billion-context(-pi)?([\/\\]|$)/.test(entry) || /(^|[/\\])billion-context(-pi)?$/.test(entry);
@@ -68801,14 +68869,14 @@ function ompConfigFile() {
         `bili plugin: PI_CODING_AGENT_DIR points at the bili overlay ${raw} \u2014 operating on the real omp home ${realHome} instead
 `
       );
-      return path17.join(realHome, "config.yml");
+      return path19.join(realHome, "config.yml");
     }
-    return path17.join(raw, "config.yml");
+    return path19.join(raw, "config.yml");
   }
-  return path17.join(os4.homedir(), ".omp", "agent", "config.yml");
+  return path19.join(os4.homedir(), ".omp", "agent", "config.yml");
 }
 function ompExtensionPath() {
-  return path17.join(selfPackageRoot(), "dist", "agent", "omp.js");
+  return path19.join(selfPackageRoot(), "dist", "agent", "omp.js");
 }
 function ompEntryValue(line) {
   return line.replace(/#.*$/, "").trim().replace(/^-\s*/, "").replace(/^["']|["']$/g, "").trim();
@@ -68838,7 +68906,7 @@ function ompInstall() {
   const file = ompConfigFile();
   const entry = ompExtensionPath();
   requireDistFile(entry);
-  fs12.mkdirSync(path17.dirname(file), { recursive: true });
+  fs12.mkdirSync(path19.dirname(file), { recursive: true });
   let text = fs12.existsSync(file) ? fs12.readFileSync(file, "utf8") : "";
   if (ompBlockLoaded(text)) return `omp: already installed (${file})`;
   {
@@ -68899,7 +68967,7 @@ function ompStatus() {
 }
 function ompPluginLoadedFrom(ompHome) {
   try {
-    return ompBlockLoaded(fs12.readFileSync(path17.join(ompHome, "config.yml"), "utf8"));
+    return ompBlockLoaded(fs12.readFileSync(path19.join(ompHome, "config.yml"), "utf8"));
   } catch {
     return false;
   }
@@ -68910,7 +68978,7 @@ function claudeMcpJson() {
 }
 function claudeInstall() {
   const root = selfPackageRoot();
-  const mcpJs = path17.join(root, "dist", "mcp.js");
+  const mcpJs = path19.join(root, "dist", "mcp.js");
   requireDistFile(mcpJs);
   const claude = process.env.CLAUDE?.trim() || "claude";
   try {
@@ -68937,14 +69005,14 @@ function claudeStatus() {
 }
 function codexToml() {
   const raw = process.env.CODEX_HOME?.trim();
-  if (raw && raw.length > 0) return path17.join(raw, "config.toml");
+  if (raw && raw.length > 0) return path19.join(raw, "config.toml");
   return homeFile(".codex/config.toml");
 }
 function codexBlock() {
   return `
 [mcp_servers.bili]
 command = ${JSON.stringify(process.execPath)}
-args = [${JSON.stringify(path17.join(selfPackageRoot(), "dist", "mcp.js"))}]
+args = [${JSON.stringify(path19.join(selfPackageRoot(), "dist", "mcp.js"))}]
 env = { BILI_MCP_PROXY = ${JSON.stringify(proxyOriginForInstall())} }
 `;
 }
@@ -68965,7 +69033,7 @@ function codexInstall() {
     const healed = malformedCodexArgs(block) ? " (repaired args: was not an array)" : "";
     return `codex: refreshed [mcp_servers.bili] -> ${file}${healed}`;
   }
-  fs12.mkdirSync(path17.dirname(file), { recursive: true });
+  fs12.mkdirSync(path19.dirname(file), { recursive: true });
   backupOnce(file);
   fs12.writeFileSync(file, text + (text.endsWith("\n") || text.length === 0 ? "" : "\n") + codexBlock());
   return `codex: installed -> ${file} [mcp_servers.bili]`;
@@ -68997,12 +69065,12 @@ function opencodeJson() {
   const raw = process.env.OPENCODE_CONFIG?.trim();
   if (raw && raw.length > 0) return raw;
   const xdg2 = process.env.XDG_CONFIG_HOME?.trim();
-  if (xdg2 && xdg2.length > 0) return path17.join(xdg2, "opencode/opencode.json");
-  return path17.join(os4.homedir(), ".config", "opencode", "opencode.json");
+  if (xdg2 && xdg2.length > 0) return path19.join(xdg2, "opencode/opencode.json");
+  return path19.join(os4.homedir(), ".config", "opencode", "opencode.json");
 }
 function opencodeInstall() {
   const file = opencodeJson();
-  const mcpJs = path17.join(selfPackageRoot(), "dist", "mcp.js");
+  const mcpJs = path19.join(selfPackageRoot(), "dist", "mcp.js");
   requireDistFile(mcpJs);
   const data = readJson(file);
   const mcp = data.mcp ?? {};
@@ -69057,11 +69125,11 @@ import { randomUUID as randomUUID5 } from "crypto";
 import fs13 from "fs";
 import net2 from "net";
 import os5 from "os";
-import path18 from "path";
+import path20 from "path";
 import { pathToFileURL } from "url";
 import { execFileSync as execFileSync3, spawn } from "child_process";
 function selfDistFile(name) {
-  return path18.join(selfPackageRoot(), "dist", name);
+  return path20.join(selfPackageRoot(), "dist", name);
 }
 var LAUNCHER_DEFAULT_HOST = "127.0.0.1";
 var LAUNCH_CLIENTS = ["pi", "codex", "claude", "omp", "opencode", "hermes", "dsh", "codebuddy", "qoder", "trae", "jcode", "kimi", "pi-test"];
@@ -69102,12 +69170,12 @@ function isLoopbackHost(host) {
   return /^127\.\d+\.\d+\.\d+$/.test(h);
 }
 function resolveCaCertPath(env) {
-  const base = env.XDG_DATA_HOME || path18.join(os5.homedir(), ".local/share");
-  return path18.join(base, "billion-context", "ca", "root-ca.pem");
+  const base = env.XDG_DATA_HOME || path20.join(os5.homedir(), ".local/share");
+  return path20.join(base, "billion-context", "ca", "root-ca.pem");
 }
 function resolveCombinedCaPath(env) {
-  const base = env.XDG_DATA_HOME || path18.join(os5.homedir(), ".local/share");
-  return path18.join(base, "billion-context", "ca", "combined-ca.pem");
+  const base = env.XDG_DATA_HOME || path20.join(os5.homedir(), ".local/share");
+  return path20.join(base, "billion-context", "ca", "combined-ca.pem");
 }
 function discoverRoutes(client, config) {
   const httpsDomains = [];
@@ -69517,7 +69585,7 @@ function prepareCodexMcpInjection(opts) {
   return { clientArgs: [], envPatch: { CODEX_HOME: overlay } };
 }
 function overlayLockPath(overlay) {
-  return path18.join(overlay, ".bili-launch.pid");
+  return path20.join(overlay, ".bili-launch.pid");
 }
 function livePidHoldsOverlay(overlay) {
   let raw;
@@ -69536,8 +69604,8 @@ function livePidHoldsOverlay(overlay) {
   return pid;
 }
 function linkOverlayEntry(realHome, overlay, entry) {
-  const target = path18.join(realHome, entry);
-  const link = path18.join(overlay, entry);
+  const target = path20.join(realHome, entry);
+  const link = path20.join(overlay, entry);
   let st2;
   try {
     st2 = fs13.lstatSync(target);
@@ -69602,7 +69670,7 @@ function mergeSqliteSet(overlay, realHome, base) {
   const members = sqliteSetMembers(base);
   const statFile = (dir, m2) => {
     try {
-      const st2 = fs13.lstatSync(path18.join(dir, m2));
+      const st2 = fs13.lstatSync(path20.join(dir, m2));
       return st2.isFile() ? st2 : void 0;
     } catch {
       return void 0;
@@ -69641,7 +69709,7 @@ function mergeSqliteSet(overlay, realHome, base) {
     undo.push(() => fs13.renameSync(dst, src));
   };
   const preserveAsConflict = (src, name) => {
-    const conflict = freeConflictName(path18.join(realHome, name));
+    const conflict = freeConflictName(path20.join(realHome, name));
     fs13.renameSync(src, conflict);
     undo.push(() => fs13.renameSync(conflict, src));
   };
@@ -69650,13 +69718,13 @@ function mergeSqliteSet(overlay, realHome, base) {
       const o = statFile(overlay, m2);
       const r = statFile(realHome, m2);
       if (winner === "overlay") {
-        if (o) movePreserving(path18.join(overlay, m2), path18.join(realHome, m2));
-        else if (r) preserveAsConflict(path18.join(realHome, m2), m2);
+        if (o) movePreserving(path20.join(overlay, m2), path20.join(realHome, m2));
+        else if (r) preserveAsConflict(path20.join(realHome, m2), m2);
       } else if (winner === "real") {
-        if (o) preserveAsConflict(path18.join(overlay, m2), m2);
+        if (o) preserveAsConflict(path20.join(overlay, m2), m2);
       } else {
-        if (o) preserveAsConflict(path18.join(overlay, m2), m2);
-        else if (r) preserveAsConflict(path18.join(realHome, m2), m2);
+        if (o) preserveAsConflict(path20.join(overlay, m2), m2);
+        else if (r) preserveAsConflict(path20.join(realHome, m2), m2);
       }
     }
     return true;
@@ -69703,10 +69771,10 @@ function refreshOverlayHome(realHome, overlay, generatedFile) {
       if (!members.some((m2) => m2 !== entry && overlayEntries.includes(m2))) continue;
       let mainSt;
       try {
-        mainSt = fs13.lstatSync(path18.join(overlay, entry));
+        mainSt = fs13.lstatSync(path20.join(overlay, entry));
       } catch {
       }
-      const keepSidecars = mainSt !== void 0 && isWriteThroughHardlink(path18.join(overlay, entry), path18.join(realHome, entry), mainSt);
+      const keepSidecars = mainSt !== void 0 && isWriteThroughHardlink(path20.join(overlay, entry), path20.join(realHome, entry), mainSt);
       dbSets.push({ base: entry, keepSidecars });
     }
     const skipEntries = /* @__PURE__ */ new Set();
@@ -69717,7 +69785,7 @@ function refreshOverlayHome(realHome, overlay, generatedFile) {
     }
     for (const entry of overlayEntries) {
       if (generatedFiles.has(entry)) continue;
-      const overlayPath = path18.join(overlay, entry);
+      const overlayPath = path20.join(overlay, entry);
       if (isGeneratedDraft(entry)) {
         try {
           fs13.unlinkSync(overlayPath);
@@ -69738,7 +69806,7 @@ function refreshOverlayHome(realHome, overlay, generatedFile) {
           target = fs13.readlinkSync(overlayPath);
         } catch {
         }
-        const wanted = realEntries.has(entry) ? path18.join(realHome, entry) : void 0;
+        const wanted = realEntries.has(entry) ? path20.join(realHome, entry) : void 0;
         if (!wanted || target !== wanted) {
           try {
             fs13.unlinkSync(overlayPath);
@@ -69746,7 +69814,7 @@ function refreshOverlayHome(realHome, overlay, generatedFile) {
           }
         }
       } else if (realEntries.has(entry)) {
-        const realPath = path18.join(realHome, entry);
+        const realPath = path20.join(realHome, entry);
         if (isWriteThroughHardlink(overlayPath, realPath, st2)) {
           try {
             fs13.unlinkSync(overlayPath);
@@ -69776,7 +69844,7 @@ function refreshOverlayHome(realHome, overlay, generatedFile) {
     for (const entry of realEntries) {
       if (generatedFiles.has(entry)) continue;
       total += 1;
-      const overlayPath = path18.join(overlay, entry);
+      const overlayPath = path20.join(overlay, entry);
       let present = false;
       try {
         fs13.lstatSync(overlayPath);
@@ -69843,7 +69911,7 @@ function mergeOverlayEntry(src, dst, excludedNames) {
     let ok = true;
     for (const entry of entries) {
       if (excludedNames?.has(entry)) continue;
-      if (!mergeOverlayEntry(path18.join(src, entry), path18.join(dst, entry), excludedNames)) ok = false;
+      if (!mergeOverlayEntry(path20.join(src, entry), path20.join(dst, entry), excludedNames)) ok = false;
     }
     return ok;
   }
@@ -69896,7 +69964,7 @@ function piPluginInstalled(piHome) {
   const root = selfPackageRoot();
   if (!root) return false;
   try {
-    const parsed = JSON.parse(fs13.readFileSync(path18.join(piHome, "settings.json"), "utf8"));
+    const parsed = JSON.parse(fs13.readFileSync(path20.join(piHome, "settings.json"), "utf8"));
     const list = Array.isArray(parsed.packages) ? parsed.packages.map(String) : [];
     return list.some((p2) => isBiliPiEntry(p2, root));
   } catch {
@@ -69904,10 +69972,10 @@ function piPluginInstalled(piHome) {
   }
 }
 function writeOverlayFileAtomic(overlay, fileName, contents) {
-  const draft = path18.join(overlay, `.${fileName}.${process.pid}.tmp`);
+  const draft = path20.join(overlay, `.${fileName}.${process.pid}.tmp`);
   try {
     fs13.writeFileSync(draft, contents);
-    fs13.renameSync(draft, path18.join(overlay, fileName));
+    fs13.renameSync(draft, path20.join(overlay, fileName));
   } catch {
     try {
       fs13.rmSync(draft, { force: true });
@@ -69916,7 +69984,7 @@ function writeOverlayFileAtomic(overlay, fileName, contents) {
   }
 }
 function prepareDshHome(dshHome, origin, rewrites) {
-  const cfgPath = path18.join(dshHome, "settings.yaml");
+  const cfgPath = path20.join(dshHome, "settings.yaml");
   let txt;
   try {
     txt = fs13.readFileSync(cfgPath, "utf8");
@@ -69968,7 +70036,7 @@ env = { BILI_MCP_PROXY = ${JSON.stringify(origin)}, BILI_CONVERSATION_ID = ${JSO
 function prepareCodexHome(codexHome, origin, conversationId2) {
   let txt = "";
   try {
-    txt = fs13.readFileSync(path18.join(codexHome, "config.toml"), "utf8");
+    txt = fs13.readFileSync(path20.join(codexHome, "config.toml"), "utf8");
   } catch {
   }
   const overlay = `${codexHome}-bili`;
@@ -69987,7 +70055,7 @@ function writeDshAcpPatch(dshHome) {
   writeOverlayFileAtomic(dir, ".bili-acp.patch.yml", `- insert:
     - name: ${pluginUrl}
 `);
-  const file = path18.join(dir, ".bili-acp.patch.yml");
+  const file = path20.join(dir, ".bili-acp.patch.yml");
   try {
     return fs13.existsSync(file) ? file : void 0;
   } catch {
@@ -70037,12 +70105,12 @@ function prepareOpencodeHttpRewrite(userRoot, origin, httpRewrites, httpsRewrite
     rewrite(httpRewrites, true);
     rewrite(httpsRewrites, false);
   }
-  const tmp = fs13.mkdtempSync(path18.join(os5.tmpdir(), "bili-opencode-"));
+  const tmp = fs13.mkdtempSync(path20.join(os5.tmpdir(), "bili-opencode-"));
   let pluginEntry = pluginPath;
   if (pluginPath && pluginDirMode) {
-    const wrapDir = path18.join(tmp, "plugin");
+    const wrapDir = path20.join(tmp, "plugin");
     fs13.mkdirSync(wrapDir);
-    fs13.writeFileSync(path18.join(wrapDir, "index.js"), `export { default } from ${JSON.stringify(pluginPath)};
+    fs13.writeFileSync(path20.join(wrapDir, "index.js"), `export { default } from ${JSON.stringify(pluginPath)};
 `);
     pluginEntry = wrapDir;
   }
@@ -70062,24 +70130,24 @@ function prepareOpencodeHttpRewrite(userRoot, origin, httpRewrites, httpsRewrite
       root[key] = root[key].map((entry) => absolutizePluginEntry(baseDir, entry));
     }
   }
-  const tmpFile = path18.join(tmp, "opencode.json");
+  const tmpFile = path20.join(tmp, "opencode.json");
   fs13.writeFileSync(tmpFile, JSON.stringify(root));
   return tmpFile;
 }
 function isRelativeLocalPluginSpec(spec) {
-  return typeof spec === "string" && spec.startsWith(".") && !path18.isAbsolute(spec);
+  return typeof spec === "string" && spec.startsWith(".") && !path20.isAbsolute(spec);
 }
 function absolutizePluginEntry(baseDir, entry) {
   if (typeof entry === "string") {
-    return isRelativeLocalPluginSpec(entry) ? path18.resolve(baseDir, entry) : entry;
+    return isRelativeLocalPluginSpec(entry) ? path20.resolve(baseDir, entry) : entry;
   }
   if (Array.isArray(entry) && entry.length > 0 && typeof entry[0] === "string") {
     if (!isRelativeLocalPluginSpec(entry[0])) return entry;
-    return [path18.resolve(baseDir, entry[0]), ...entry.slice(1)];
+    return [path20.resolve(baseDir, entry[0]), ...entry.slice(1)];
   }
   if (entry !== null && typeof entry === "object" && !Array.isArray(entry) && "package" in entry) {
     const obj = entry;
-    if (isRelativeLocalPluginSpec(obj.package)) return { ...obj, package: path18.resolve(baseDir, obj.package) };
+    if (isRelativeLocalPluginSpec(obj.package)) return { ...obj, package: path20.resolve(baseDir, obj.package) };
   }
   return entry;
 }
@@ -70225,7 +70293,7 @@ async function ensureProxyRunning(opts, deps = {}) {
   const port = opts.port > 0 ? opts.port : await pickEphemeralPort(opts.host);
   const script = process.argv[1];
   if (!script) throw new Error("bili: cannot resolve launcher script path");
-  const logPath2 = path18.join(os5.tmpdir(), `bili-proxy-${port}.log`);
+  const logPath2 = path20.join(os5.tmpdir(), `bili-proxy-${port}.log`);
   const logFd = fs13.openSync(logPath2, "a");
   const claimMarker = () => claimStartingMarker({ token: launchToken, pid: process.pid, host: opts.host, port, startedAt: now() });
   let claimed = claimMarker();
@@ -70329,7 +70397,7 @@ function planClientSpawn(cmd, args, env, platform = process.platform) {
   if (platform !== "win32") return { command: cmd, args: [...args] };
   const lower = cmd.toLowerCase();
   const base = cmd.slice(Math.max(cmd.lastIndexOf("/"), cmd.lastIndexOf("\\")) + 1);
-  const needsCmd = lower.endsWith(".cmd") || lower.endsWith(".bat") || !path18.extname(base);
+  const needsCmd = lower.endsWith(".cmd") || lower.endsWith(".bat") || !path20.extname(base);
   if (!needsCmd) return { command: cmd, args: [...args] };
   const comspec = nonEmpty2(env.COMSPEC) ? env.COMSPEC : "cmd.exe";
   return {
@@ -70355,10 +70423,10 @@ var PATH_EXTS = process.platform === "win32" ? [".cmd", ".bat", ".exe", ""] : ["
 function resolveOnPath(name, env) {
   const p2 = env.PATH;
   if (!p2) return void 0;
-  for (const dir of p2.split(path18.delimiter)) {
+  for (const dir of p2.split(path20.delimiter)) {
     if (!dir) continue;
     for (const ext of PATH_EXTS) {
-      const f2 = path18.join(dir, name + ext);
+      const f2 = path20.join(dir, name + ext);
       try {
         if (fs13.existsSync(f2) && fs13.statSync(f2).isFile()) return f2;
       } catch {
@@ -70378,7 +70446,7 @@ function resolveClientCommand(client, env) {
     if (piBin) return { command: piBin, prefixArgs: [] };
     const piResolved = resolveOnPath("pi", env);
     if (piResolved) return { command: piResolved, prefixArgs: [] };
-    const cli = path18.join(
+    const cli = path20.join(
       os5.homedir(),
       ".pi/agent/npm/node_modules/@earendil-works/pi-coding-agent/dist/cli.js"
     );
@@ -70399,7 +70467,7 @@ function resolveClientCommand(client, env) {
   if (client === "kimi") {
     const resolved2 = resolveOnPath("kimi", env);
     if (resolved2) return { command: resolved2, prefixArgs: [] };
-    return { command: path18.join(resolveKimiHome(env), "bin", "kimi"), prefixArgs: [] };
+    return { command: path20.join(resolveKimiHome(env), "bin", "kimi"), prefixArgs: [] };
   }
   const resolved = resolveOnPath(client, env);
   return { command: resolved ?? client, prefixArgs: [] };
@@ -70639,7 +70707,7 @@ async function runLaunch(params, deps = {}) {
       console.error(`bili: claude budget aligned \u2014 CLAUDE_CODE_AUTO_COMPACT_WINDOW=${claudeBudget.CLAUDE_CODE_AUTO_COMPACT_WINDOW}`);
     }
     if (injectMcp) {
-      const mcpFile = path18.join(os5.tmpdir(), `bili-mcp-${Date.now()}.json`);
+      const mcpFile = path20.join(os5.tmpdir(), `bili-mcp-${Date.now()}.json`);
       fs13.writeFileSync(mcpFile, JSON.stringify(buildMcpConfig(origin)));
       tmpFiles.push(mcpFile);
       clientArgs = ["--mcp-config", mcpFile, ...clientArgs];
@@ -70659,7 +70727,7 @@ async function runLaunch(params, deps = {}) {
     stopProxy(handle2);
     if (opencodeTmpFile) {
       try {
-        fs13.rmSync(path18.dirname(opencodeTmpFile), { recursive: true, force: true });
+        fs13.rmSync(path20.dirname(opencodeTmpFile), { recursive: true, force: true });
       } catch {
       }
     }
@@ -70691,7 +70759,7 @@ async function runTestPi(params, deps = {}) {
   }
   const ca = resolveCaCertPath(process.env);
   const env = buildPiEnv(handle2.origin, ca, process.env);
-  const sessionDir = path18.join(os5.tmpdir(), `bili-pi-test-${Date.now()}`);
+  const sessionDir = path20.join(os5.tmpdir(), `bili-pi-test-${Date.now()}`);
   fs13.mkdirSync(sessionDir, { recursive: true });
   const args = [
     "-p",
@@ -70720,7 +70788,7 @@ async function runTestPi(params, deps = {}) {
 
 // src/export.ts
 import { mkdirSync as mkdirSync6, writeFileSync as writeFileSync5 } from "fs";
-import path19 from "path";
+import path21 from "path";
 function fmtDate(ms2) {
   return ms2 ? new Date(ms2).toISOString().replace("T", " ").slice(0, 19) + " UTC" : "\u2014";
 }
@@ -70748,16 +70816,11 @@ function renderHandoff2(s3, full) {
   const messages = s3.lastMessages;
   if (messages && messages.length > 0) {
     const folded = s3.lastMessagesFolded === true;
-    const blocksFull = full && folded ? s3.state.blocks.filter((b2) => b2.active).map((b2) => {
-      const content = s3.blockContents.get(b2.blockId);
-      return content ? { blockId: b2.blockId, topic: b2.topic, count: content.full.count, fullText: content.full.text } : void 0;
-    }).filter((x) => x !== void 0) : void 0;
-    return renderHandoff({
+    const base = renderHandoff({
       coreMessages: messages,
       state: s3.state,
       full,
       folded,
-      blocksFull,
       meta: {
         title: s3.meta.title,
         label: s3.meta.label,
@@ -70770,6 +70833,7 @@ function renderHandoff2(s3, full) {
         ]
       }
     });
+    return base + blockSummariesSection(s3, full && folded, base);
   }
   const lines = [];
   lines.push(`# billion-context session handoff`);
@@ -70811,6 +70875,36 @@ function renderHandoff2(s3, full) {
   }
   return lines.join("\n");
 }
+function blockSummariesSection(s3, includeOriginals, priorDoc) {
+  const active = s3.state.blocks.filter((b2) => b2.active);
+  if (active.length === 0) return "";
+  const lines = [];
+  lines.push(`## Compressed block summaries`);
+  lines.push("");
+  for (const b2 of active) {
+    lines.push(`### Block ${b2.blockId}${b2.topic ? ` \u2014 ${b2.topic}` : ""}`);
+    lines.push("");
+    lines.push(`tier ${b2.tier} \xB7 ~${b2.compressedTokens} tokens compressed \xB7 ${fmtDate(b2.createdAt)}`);
+    lines.push("");
+    const summary = b2.summary.trim();
+    if (summary === "") {
+      lines.push("_no summary recorded_");
+    } else if (priorDoc.includes(summary)) {
+      lines.push("_summary already shown in the conversation view above_");
+    } else {
+      lines.push(summary);
+    }
+    lines.push("");
+    const content = includeOriginals ? s3.blockContents.get(b2.blockId) : void 0;
+    if (content) {
+      lines.push(`#### Original messages (${content.full.count})`);
+      lines.push("");
+      lines.push(content.full.text.trim());
+      lines.push("");
+    }
+  }
+  return lines.join("\n");
+}
 function matchSession2(sessions2, selector) {
   return matchSession(sessions2, selector, (s3) => s3.meta.label);
 }
@@ -70837,7 +70931,7 @@ async function exportSession(selector, opts = {}) {
   }
   const markdown = renderHandoff2(matches[0], opts.full ?? false);
   if (opts.output) {
-    mkdirSync6(path19.dirname(path19.resolve(opts.output)), { recursive: true });
+    mkdirSync6(path21.dirname(path21.resolve(opts.output)), { recursive: true });
     writeFileSync5(opts.output, markdown, "utf8");
     return `written to ${opts.output}`;
   }