billion-context 0.1.49 → 0.1.50-pr.225.44

package/dist/index.js

@@ -47739,6 +47739,75 @@ async function fetchWithTimeout(url, opts, timeoutMs = UPSTREAM_TIMEOUT_MS, exte
     throw e;
   }
 }
+var UpstreamHttpError = class extends Error {
+  status;
+  body;
+  attempts;
+  constructor(status, body, attempts) {
+    super(`upstream error ${status}`);
+    this.name = "UpstreamHttpError";
+    this.status = status;
+    this.body = body;
+    this.attempts = attempts;
+  }
+};
+var TRANSIENT_BODY_MARKERS = [
+  "captcha",
+  "verify failed",
+  "risk control",
+  "\u98CE\u63A7",
+  "rate limit",
+  "too many requests",
+  "try again"
+];
+function isTransientUpstreamError(status, body) {
+  if (status === 429 || status >= 500) return true;
+  if (status < 400) return false;
+  const lower = body.toLowerCase();
+  return TRANSIENT_BODY_MARKERS.some((marker) => lower.includes(marker));
+}
+var REPLAY_MAX_ATTEMPTS = 3;
+function replayMaxAttempts() {
+  const raw = Number(process.env.BILI_REPLAY_RETRY_MAX);
+  return Number.isInteger(raw) && raw >= 1 ? raw : REPLAY_MAX_ATTEMPTS;
+}
+function replayBaseDelayMs() {
+  const raw = Number(process.env.BILI_REPLAY_RETRY_BASE_MS);
+  return Number.isFinite(raw) && raw >= 0 ? raw : 1500;
+}
+function replayBackoffMs(attempt) {
+  return replayBaseDelayMs() * 2 ** (attempt - 1);
+}
+function sleep(ms2, signal) {
+  if (ms2 <= 0 || signal?.aborted) return Promise.resolve();
+  return new Promise((resolve) => {
+    let timer2 = null;
+    const finish2 = () => {
+      if (timer2) clearTimeout(timer2);
+      if (signal) signal.removeEventListener("abort", finish2);
+      resolve();
+    };
+    timer2 = setTimeout(finish2, ms2);
+    if (signal) signal.addEventListener("abort", finish2, { once: true });
+  });
+}
+async function fetchWithRetry(url, opts, timeoutMs, externalSignal, onRetry) {
+  const maxAttempts = replayMaxAttempts();
+  for (let attempt = 1; ; attempt++) {
+    const result = await fetchWithTimeout(url, opts, timeoutMs, externalSignal);
+    if (result.response.ok) return result;
+    const errText2 = await result.response.text().catch(() => "upstream error");
+    result.clearTimer();
+    const lastAttempt = attempt >= maxAttempts;
+    if (!lastAttempt && isTransientUpstreamError(result.response.status, errText2)) {
+      const delayMs = replayBackoffMs(attempt);
+      onRetry?.({ attempt, status: result.response.status, detail: errText2, delayMs, maxAttempts });
+      await sleep(delayMs, externalSignal);
+      continue;
+    }
+    throw new UpstreamHttpError(result.response.status, errText2, attempt);
+  }
+}
 
 // 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" }] } } };
@@ -49834,6 +49903,12 @@ ${icon} [ACP] ${inner}
 
 // src/loop/core.ts
 var MAX_LOOP_ROUNDS = 10;
+function isLoopThinking(m2) {
+  return m2.contentType === "reasoning" && typeof m2.id === "string" && m2.id.startsWith("acp_loop_");
+}
+function stripLoopThinking(messages) {
+  return messages.filter((m2) => !isLoopThinking(m2));
+}
 function executeProxyTool(toolName, args, ctx, callId) {
   if (toolName === "compress") {
     return applyRanges(parseCompressInput(args, callId), ctx);
@@ -49907,6 +49982,7 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
   let activeClearTimer = null;
   let currentUpstream = upstream;
   const coreMessages = [...ctx.messages];
+  let degradedRetried = false;
   try {
     for (let round = 1; round <= MAX_LOOP_ROUNDS; round++) {
       if (signal?.aborted) break;
@@ -49917,6 +49993,7 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
       const calls = [];
       let usage = {};
       let finishReason;
+      let suppressCompletion = false;
       for await (const ev of adapter.parseStream(currentUpstream, round)) {
         if (signal?.aborted) break;
         if (ev.kind === "text") {
@@ -49954,6 +50031,7 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
           };
         } else if (ev.kind === "done") {
           finishReason = ev.finishReason;
+          suppressCompletion = ev.suppressCompletion === true;
         } else if (ev.kind === "meta") {
           if (round === 1 || !ev.firstRoundOnly) {
             yield ev.chunk;
@@ -50009,7 +50087,7 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
         if (reasoningSegments.length > 0) {
           for (let i = 0; i < reasoningSegments.length; i++) {
             const seg = reasoningSegments[i];
-            if (seg.text.length === 0 && seg.signature.length === 0) continue;
+            if (seg.text.length === 0 || seg.signature.length === 0) continue;
             const reasoningMsg = {
               id: i === 0 ? `acp_loop_r${round}_reasoning` : `acp_loop_r${round}_reasoning_${i + 1}`,
               role: "assistant",
@@ -50073,7 +50151,9 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
       }
       const reRequest = proxyResults.length > 0 && realCalls === 0;
       if (!reRequest) {
-        yield adapter.emitCompletion({ finishReason, usage });
+        if (!suppressCompletion) {
+          yield adapter.emitCompletion({ finishReason, usage });
+        }
         return;
       }
       if (round >= MAX_LOOP_ROUNDS) {
@@ -50084,7 +50164,22 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
       }
       ctx.log(`[acp-loop] round ${round}: proxy tool executed; re-requesting so the model sees the result`);
       if (signal?.aborted) break;
-      const newBody = adapter.buildRequest(coreMessages, systemPrompt, requestBody);
+      const fetchUpstream = (body) => fetchWithRetry(
+        requestOptions.url,
+        {
+          method: "POST",
+          headers: requestOptions.headers,
+          body: JSON.stringify(body),
+          ...ctx.proxyUrl ? { dispatcher: proxyDispatcher(ctx.proxyUrl) } : {}
+        },
+        void 0,
+        signal,
+        (info) => {
+          ctx.log(`[acp-proxy: upstream rejected replay (HTTP ${info.status}: ${info.detail.slice(0, 120)}); likely provider risk-control \u2014 retrying in ${info.delayMs}ms (attempt ${info.attempt}/${info.maxAttempts})]`);
+          log("warn", `[acp-loop] upstream rejected replay (HTTP ${info.status}); retrying in ${info.delayMs}ms (attempt ${info.attempt}/${info.maxAttempts})`);
+        }
+      );
+      let newBody = adapter.buildRequest(coreMessages, systemPrompt, requestBody);
       if (process.env.ACP_DUMP_REQ !== "0" && ctx.debug) {
         try {
           const fs11 = await import("fs");
@@ -50095,28 +50190,39 @@ async function* runCompressLoop(upstream, ctx, requestBody, requestOptions, adap
         } catch {
         }
       }
-      const { response: resp, clearTimer } = await fetchWithTimeout(
-        requestOptions.url,
-        {
-          method: "POST",
-          headers: requestOptions.headers,
-          body: JSON.stringify(newBody),
-          ...ctx.proxyUrl ? { dispatcher: proxyDispatcher(ctx.proxyUrl) } : {}
-        },
-        void 0,
-        signal
-      );
-      if (!resp.ok || !resp.body) {
-        clearTimer();
-        const errText2 = await resp.text().catch(() => "upstream error");
-        ctx.log(`[acp-proxy: compress loop upstream error ${resp.status}: ${errText2.slice(0, 200)}]`);
-        log("error", `[acp-loop] upstream error ${resp.status}: ${errText2.slice(0, 200)}`);
-        yield adapter.emitError(`upstream error ${resp.status}: ${errText2.slice(0, 200)}`);
+      let respResult;
+      try {
+        try {
+          respResult = await fetchUpstream(newBody);
+        } catch (e) {
+          if (!(e instanceof UpstreamHttpError) || e.status < 400 || e.status >= 500 || degradedRetried || !coreMessages.some(isLoopThinking)) {
+            throw e;
+          }
+          degradedRetried = true;
+          const stripped = stripLoopThinking(coreMessages);
+          coreMessages.length = 0;
+          coreMessages.push(...stripped);
+          ctx.log(`[acp-loop] round ${round}: re-request rejected (${e.status}: ${e.body.slice(0, 200)}); retrying without replayed thinking blocks`);
+          log("warn", `[acp-loop] re-request rejected (${e.status}); retrying without thinking replay: ${e.body.slice(0, 200)}`);
+          newBody = adapter.buildRequest(coreMessages, systemPrompt, requestBody);
+          respResult = await fetchUpstream(newBody);
+        }
+      } catch (e) {
+        if (!(e instanceof UpstreamHttpError)) throw e;
+        const suffix = e.attempts > 1 ? ` after ${e.attempts} attempt(s)` : "";
+        ctx.log(`[acp-proxy: compress loop upstream error ${e.status}${suffix}: ${e.body.slice(0, 200)}]`);
+        log("error", `[acp-loop] upstream error ${e.status}${suffix}: ${e.body.slice(0, 200)}`);
+        yield adapter.emitError(`upstream error ${e.status}${suffix}: ${e.body.slice(0, 200)}`);
         return;
       }
-      currentUpstream = resp.body;
+      if (!respResult.response.body) {
+        respResult.clearTimer();
+        yield adapter.emitError(`upstream error ${respResult.response.status}: empty response body`);
+        return;
+      }
+      currentUpstream = respResult.response.body;
       if (activeClearTimer) activeClearTimer();
-      activeClearTimer = clearTimer;
+      activeClearTimer = respResult.clearTimer;
     }
   } finally {
     if (activeClearTimer) {
@@ -50554,6 +50660,31 @@ var PROXY_TOOL_SET = /* @__PURE__ */ new Set([
   "bili_search_context",
   "bili_status"
 ]);
+function filterRealToolFragments(parsed, realIndexes) {
+  const choices = parsed.choices;
+  const choice = choices?.[0];
+  const delta = choice?.delta;
+  const tcs = delta?.tool_calls;
+  if (!tcs || tcs.length === 0) return "keep";
+  let anyReal = false;
+  let anyProxy = false;
+  for (const tc of tcs) {
+    const idx = typeof tc.index === "number" ? tc.index : 0;
+    if (realIndexes.has(idx)) anyReal = true;
+    else anyProxy = true;
+  }
+  if (!anyProxy) return "keep";
+  if (!anyReal) return null;
+  const filtered = { ...parsed };
+  const filteredChoices = [...choices];
+  const filteredChoice = { ...choice };
+  const filteredDelta = { ...delta };
+  filteredDelta.tool_calls = tcs.filter((tc) => realIndexes.has(typeof tc.index === "number" ? tc.index : 0));
+  filteredChoice.delta = filteredDelta;
+  filteredChoices[0] = filteredChoice;
+  filtered.choices = filteredChoices;
+  return filtered;
+}
 async function* iterSseChunks(stream2) {
   const reader = stream2.getReader();
   let buf = "";
@@ -50656,7 +50787,62 @@ function createOpenaiAdapter(requestBody, clientSystem) {
     },
     async *parseStream(upstream, _round) {
       const pending = /* @__PURE__ */ new Map();
+      const rawToolChunks = [];
       let sawRealToolCall = false;
+      const flushPendingAsStructured = function* () {
+        for (const [, tc] of pending) {
+          if (tc.name.length > 0 || tc.id.length > 0) {
+            yield {
+              kind: "tool_call",
+              name: tc.name,
+              callId: tc.id,
+              arguments: tc.arguments
+            };
+          }
+        }
+        pending.clear();
+      };
+      const settleToolCalls = function* () {
+        const realIndexes = /* @__PURE__ */ new Set();
+        for (const [idx, tc] of pending) {
+          if (tc.name.length > 0 && !PROXY_TOOL_SET.has(tc.name)) realIndexes.add(idx);
+        }
+        sawRealToolCall = realIndexes.size > 0;
+        if (!sawRealToolCall) {
+          yield* flushPendingAsStructured();
+          return;
+        }
+        for (const [idx, tc] of pending) {
+          if (realIndexes.has(idx) && (tc.name.length > 0 || tc.id.length > 0)) {
+            yield {
+              kind: "tool_call",
+              name: tc.name,
+              callId: tc.id,
+              arguments: tc.arguments,
+              passthrough: true
+            };
+          }
+        }
+        for (const { json, parsed } of rawToolChunks) {
+          const filtered = filterRealToolFragments(parsed, realIndexes);
+          if (filtered === "keep") {
+            yield { kind: "meta", chunk: Buffer.from("data: " + json + "\n\n", "utf8") };
+          } else if (filtered !== null) {
+            yield { kind: "meta", chunk: Buffer.from("data: " + JSON.stringify(filtered) + "\n\n", "utf8") };
+          }
+        }
+        for (const [idx, tc] of pending) {
+          if (!realIndexes.has(idx) && tc.name.length > 0) {
+            yield {
+              kind: "tool_call",
+              name: tc.name,
+              callId: tc.id,
+              arguments: tc.arguments
+            };
+          }
+        }
+        pending.clear();
+      };
       for await (const eventStr of iterSseChunks(upstream)) {
         const dataLine = eventStr.split("\n").find((l) => l.startsWith("data:"));
         if (!dataLine) continue;
@@ -50666,18 +50852,11 @@ function createOpenaiAdapter(requestBody, clientSystem) {
             yield { kind: "meta", chunk: Buffer.from(eventStr + "\n\n", "utf8") };
             continue;
           }
-          for (const [, tc] of pending) {
-            if (tc.name.length > 0 || tc.id.length > 0) {
-              yield {
-                kind: "tool_call",
-                name: tc.name,
-                callId: tc.id,
-                arguments: tc.arguments
-              };
-            }
+          yield* settleToolCalls();
+          if (sawRealToolCall) {
+            yield { kind: "meta", chunk: Buffer.from(eventStr + "\n\n", "utf8") };
           }
-          pending.clear();
-          yield { kind: "done", finishReason: "stop" };
+          yield { kind: "done", finishReason: "stop", ...sawRealToolCall ? { suppressCompletion: true } : {} };
           continue;
         }
         let parsed;
@@ -50705,32 +50884,8 @@ function createOpenaiAdapter(requestBody, clientSystem) {
         const delta = choice.delta;
         const finishReason = typeof choice.finish_reason === "string" ? choice.finish_reason : void 0;
         if (finishReason) {
-          if (sawRealToolCall) {
-            const u22 = parsed.usage;
-            const pd2 = u22?.prompt_tokens_details;
-            yield {
-              kind: "usage",
-              inputTokens: typeof u22?.prompt_tokens === "number" ? u22.prompt_tokens : void 0,
-              outputTokens: typeof u22?.completion_tokens === "number" ? u22.completion_tokens : void 0,
-              cachedTokens: typeof pd2?.cached_tokens === "number" ? pd2.cached_tokens : void 0
-            };
-            yield { kind: "meta", chunk: rawBuf };
-            yield { kind: "done", finishReason };
-            continue;
-          }
-          let yieldedToolCall = false;
-          for (const [, tc] of pending) {
-            if (tc.name.length > 0 || tc.id.length > 0) {
-              yieldedToolCall = true;
-              yield {
-                kind: "tool_call",
-                name: tc.name,
-                callId: tc.id,
-                arguments: tc.arguments
-              };
-            }
-          }
-          pending.clear();
+          const hadToolCalls = [...pending.values()].some((tc) => tc.name.length > 0 || tc.id.length > 0);
+          yield* settleToolCalls();
           const u2 = parsed.usage;
           const pd = u2?.prompt_tokens_details;
           yield {
@@ -50739,10 +50894,15 @@ function createOpenaiAdapter(requestBody, clientSystem) {
             outputTokens: typeof u2?.completion_tokens === "number" ? u2.completion_tokens : void 0,
             cachedTokens: typeof pd?.cached_tokens === "number" ? pd.cached_tokens : void 0
           };
-          yield {
-            kind: "done",
-            finishReason: yieldedToolCall && finishReason === "stop" ? "tool_calls" : finishReason
-          };
+          if (sawRealToolCall) {
+            yield { kind: "meta", chunk: rawBuf };
+            yield { kind: "done", finishReason, suppressCompletion: true };
+          } else {
+            yield {
+              kind: "done",
+              finishReason: hadToolCalls && finishReason === "stop" ? "tool_calls" : finishReason
+            };
+          }
         }
         if (!delta) continue;
         if (typeof delta.reasoning_content === "string" && delta.reasoning_content.length > 0) {
@@ -50750,22 +50910,7 @@ function createOpenaiAdapter(requestBody, clientSystem) {
         }
         if (delta.tool_calls) {
           const tcs = delta.tool_calls;
-          let allProxy = true;
-          for (const tc of tcs) {
-            const fn = tc.function;
-            const name = typeof fn?.name === "string" ? fn.name : "";
-            if (!PROXY_TOOL_SET.has(name)) {
-              allProxy = false;
-            }
-          }
-          if (!allProxy) {
-            sawRealToolCall = true;
-            yield { kind: "meta", chunk: rawBuf };
-            if (typeof delta.content === "string" && delta.content.length > 0) {
-              yield { kind: "text", delta: delta.content };
-            }
-            continue;
-          }
+          rawToolChunks.push({ json: jsonStr, parsed });
           for (const tc of tcs) {
             const idx = typeof tc.index === "number" ? tc.index : 0;
             const fn = tc.function;
@@ -50778,7 +50923,7 @@ function createOpenaiAdapter(requestBody, clientSystem) {
               pending.set(idx, buf);
             } else {
               if (id) buf.id = id;
-              if (name) buf.name = name;
+              buf.name += name;
               buf.arguments += args;
             }
           }
@@ -51273,25 +51418,31 @@ async function compressLoopResponsesJson(initialResponse, ctx, requestBody, requ
       } catch (error) {
         log("warn", `[acp-compress-args] ${call.name} JSON.parse failed: ${String(error)}`);
       }
-      const result = executeProxyTool2(call.name, args, ctx);
-      ctx.log(`[acp-proxy: responses JSON ${call.name} \u2192 ${result.slice(0, 120).replace(/\n/g, " ")}]`);
-      inputItems.push({ type: "message", role: "developer", content: buildVisibilityMarker(call.name, result) });
+      const result2 = executeProxyTool2(call.name, args, ctx);
+      ctx.log(`[acp-proxy: responses JSON ${call.name} \u2192 ${result2.slice(0, 120).replace(/\n/g, " ")}]`);
+      inputItems.push({ type: "message", role: "developer", content: buildVisibilityMarker(call.name, result2) });
     }
     requestBody.input = inputItems;
-    const { response, clearTimer } = await fetchWithTimeout(requestOptions.url, {
+    const result = await fetchWithRetry(requestOptions.url, {
       method: "POST",
       headers: requestOptions.headers,
       body: JSON.stringify(requestBody),
       ...ctx.proxyUrl ? { dispatcher: proxyDispatcher(ctx.proxyUrl) } : {}
+    }, void 0, void 0, (info) => {
+      ctx.log(`[acp-proxy: responses upstream rejected replay (HTTP ${info.status}: ${info.detail.slice(0, 120)}); likely provider risk-control \u2014 retrying in ${info.delayMs}ms (attempt ${info.attempt}/${info.maxAttempts})]`);
+      log("warn", `[acp-compress-responses] upstream rejected replay (HTTP ${info.status}); retrying in ${info.delayMs}ms (attempt ${info.attempt}/${info.maxAttempts})`);
+    }).catch((e) => {
+      if (e instanceof UpstreamHttpError) {
+        const suffix = e.attempts > 1 ? ` after ${e.attempts} attempt(s)` : "";
+        ctx.log(`[acp-proxy: responses compress loop upstream error ${e.status}${suffix}: ${e.body.slice(0, 200)}]`);
+        log("error", `[acp-compress-responses] upstream error ${e.status}${suffix}: ${e.body.slice(0, 200)}`);
+      }
+      throw e;
     });
     try {
-      if (!response.ok) {
-        const detail = await response.text().catch(() => "upstream error");
-        throw new Error(`responses compress loop upstream error ${response.status}: ${detail.slice(0, 200)}`);
-      }
-      current = await response.json();
+      current = await result.response.json();
     } finally {
-      clearTimer();
+      result.clearTimer();
     }
   }
   ctx.log(`[acp-proxy: responses JSON compress loop limit (${MAX_LOOP_ROUNDS}) reached]`);
@@ -52232,6 +52383,10 @@ function resolveOmpHome(env) {
   const h = os.homedir();
   return nonEmpty2(env.PI_CODING_AGENT_DIR) ? env.PI_CODING_AGENT_DIR : path8.join(h, ".omp", "agent");
 }
+function resolveHermesHome(env) {
+  const h = os.homedir();
+  return nonEmpty2(env.HERMES_HOME) ? env.HERMES_HOME : path8.join(h, ".hermes");
+}
 function readClaudeSettings(homeDir, cwd, env = process.env) {
   const files = [
     path8.join(homeDir, ".claude", "settings.json"),
@@ -52343,6 +52498,82 @@ function readOmpConfig(ompHome) {
   }
   return parseOmpYaml(text);
 }
+function parseHermesYaml(text) {
+  const result = { providers: {} };
+  let mode = "none";
+  let sectionIndent = -1;
+  let entryIndent = -1;
+  let current = null;
+  let anonCount = 0;
+  for (const rawLine of text.split(/\r?\n/)) {
+    const trimmed = rawLine.trim();
+    if (!trimmed || trimmed.startsWith("#")) continue;
+    const indent = rawLine.length - rawLine.trimStart().length;
+    if (mode === "none") {
+      if (/^providers:\s*(#.*)?$/.test(trimmed)) {
+        mode = "dict";
+        sectionIndent = indent;
+        entryIndent = -1;
+        current = null;
+      } else if (/^custom_providers:\s*(#.*)?$/.test(trimmed)) {
+        mode = "list";
+        sectionIndent = indent;
+        entryIndent = -1;
+        current = null;
+      }
+      continue;
+    }
+    if (indent <= sectionIndent) {
+      mode = "none";
+      sectionIndent = -1;
+      entryIndent = -1;
+      current = null;
+      if (/^providers:\s*(#.*)?$/.test(trimmed)) {
+        mode = "dict";
+        sectionIndent = indent;
+      } else if (/^custom_providers:\s*(#.*)?$/.test(trimmed)) {
+        mode = "list";
+        sectionIndent = indent;
+      }
+      continue;
+    }
+    if (mode === "dict") {
+      if (entryIndent === -1) entryIndent = indent;
+      if (indent === entryIndent) {
+        const m2 = /^([A-Za-z0-9_.-]+):/.exec(trimmed);
+        current = m2 ? m2[1] : null;
+        if (current && !result.providers[current]) result.providers[current] = {};
+      } else if (indent > entryIndent && current) {
+        const apiMatch = /^api:\s*(\S+)/.exec(trimmed);
+        if (apiMatch) result.providers[current].api = apiMatch[1];
+      }
+    } else {
+      const dashMatch = /^-\s+(.*)$/.exec(trimmed);
+      if (dashMatch) {
+        entryIndent = indent;
+        const nameMatch = /name:\s*([A-Za-z0-9_.-]+)/.exec(dashMatch[1]);
+        current = nameMatch ? nameMatch[1] : `custom-${++anonCount}`;
+        if (!result.providers[current]) result.providers[current] = {};
+        const inlineUrl = /^(?:base_url|api|url):\s*(\S+)/.exec(dashMatch[1]);
+        if (inlineUrl) result.providers[current].api = inlineUrl[1];
+      } else if (current) {
+        const urlMatch = /^(?:base_url|api|url):\s*(\S+)/.exec(trimmed);
+        if (urlMatch) result.providers[current].api = urlMatch[1];
+      }
+    }
+  }
+  return result;
+}
+function readHermesConfig(hermesHome) {
+  const cfgPath = path8.join(hermesHome, "config.yaml");
+  let text;
+  try {
+    text = fs4.readFileSync(cfgPath, "utf8");
+  } catch {
+    return { providers: {} };
+  }
+  return parseHermesYaml(text);
+}
 function resolveOpencodeConfigFile(env) {
   if (nonEmpty2(env.OPENCODE_CONFIG)) return env.OPENCODE_CONFIG;
   const xdg2 = nonEmpty2(env.XDG_CONFIG_HOME) ? env.XDG_CONFIG_HOME : path8.join(os.homedir(), ".config");
@@ -52421,6 +52652,7 @@ function loadClientConfig(env, cwd) {
   config.zcode = readZcodeConfig(zcodeHome);
   config.omp = readOmpConfig(resolveOmpHome(env));
   config.opencode = readOpencodeConfig(resolveOpencodeConfigFile(env));
+  config.hermes = readHermesConfig(resolveHermesHome(env));
   return config;
 }
 
@@ -58513,7 +58745,7 @@ function selfDistFile(name) {
 }
 var LAUNCHER_DEFAULT_HOST = "127.0.0.1";
 var LAUNCHER_DEFAULT_PORT = 8787;
-var LAUNCH_CLIENTS = ["pi", "codex", "claude", "omp", "opencode", "pi-test"];
+var LAUNCH_CLIENTS = ["pi", "codex", "claude", "omp", "opencode", "hermes", "pi-test"];
 var HEALTH_PATH = "/__bili/health";
 var HEALTH_POLL_INTERVAL_MS = 200;
 var SPAWN_WAIT_MS = 2e4;
@@ -58607,6 +58839,21 @@ function discoverRoutes(client, config) {
     for (const [name, prov] of Object.entries(config.opencode?.providers ?? {})) {
       classify(prov.baseURL, name);
     }
+  } else if (client === "hermes") {
+    const hermesSeen = /* @__PURE__ */ new Set();
+    for (const [name, prov] of Object.entries(config.hermes?.providers ?? {})) {
+      if (!nonEmpty2(prov.api)) continue;
+      const real = unwrapUpstream2(prov.api);
+      try {
+        const url = new URL(real);
+        if (url.protocol !== "http:" && url.protocol !== "https:") continue;
+        if (hermesSeen.has(name)) continue;
+        hermesSeen.add(name);
+        rewriteKeys.add(name);
+        httpRewrites.push({ key: name, realUpstream: real });
+      } catch {
+      }
+    }
   } else {
     for (const [name, prov] of Object.entries(config.codex?.providers ?? {})) {
       classify(prov.baseUrl, `model_providers.${name}.base_url`);
@@ -58647,7 +58894,7 @@ function launcherDirectUrl(env) {
   return env.BILI_LAUNCHER_DIRECT === "1";
 }
 function launcherInjectMcp(env, base) {
-  return base !== "pi" && base !== "omp" && base !== "opencode" && env.BILI_LAUNCHER_PLUGIN === "1";
+  return base !== "pi" && base !== "omp" && base !== "opencode" && base !== "hermes" && env.BILI_LAUNCHER_PLUGIN === "1";
 }
 function buildMcpConfig(origin) {
   const script = selfDistFile("mcp.js");
@@ -58785,6 +59032,45 @@ function prepareOmpHttpRewrite(ompHome, origin, httpRewrites, httpsRewrites) {
   fs10.writeFileSync(path13.join(tmp, "models.yml"), lines.join("\n"));
   return tmp;
 }
+function prepareHermesHome(hermesHome, origin, rewrites) {
+  if (rewrites.length === 0) return void 0;
+  const cfgPath = path13.join(hermesHome, "config.yaml");
+  let txt;
+  try {
+    txt = fs10.readFileSync(cfgPath, "utf8");
+  } catch {
+    return void 0;
+  }
+  const wrapSet = new Set(rewrites.map((r) => r.realUpstream));
+  const eol = txt.includes("\r\n") ? "\r\n" : "\n";
+  const lines = txt.split(/\r?\n/);
+  let changed = false;
+  for (let i = 0; i < lines.length; i++) {
+    const rawLine = lines[i];
+    const m2 = /^(\s*(?:api|base_url|url):\s*)(\S+)(\s+#.*)?$/.exec(rawLine);
+    if (!m2) continue;
+    const rawUrl = m2[2];
+    if (!/^https?:\/\//i.test(rawUrl)) continue;
+    const real = unwrapUpstream2(rawUrl);
+    if (!wrapSet.has(real)) continue;
+    lines[i] = `${m2[1]}${wrapUpstream(origin, real)}${m2[3] ?? ""}`;
+    changed = true;
+  }
+  if (!changed) return void 0;
+  const tmp = fs10.mkdtempSync(path13.join(os5.tmpdir(), "bili-hermes-"));
+  try {
+    for (const entry of fs10.readdirSync(hermesHome)) {
+      if (entry === "config.yaml") continue;
+      try {
+        fs10.symlinkSync(path13.join(hermesHome, entry), path13.join(tmp, entry));
+      } catch {
+      }
+    }
+  } catch {
+  }
+  fs10.writeFileSync(path13.join(tmp, "config.yaml"), lines.join(eol));
+  return tmp;
+}
 function prepareOpencodeHttpRewrite(configFile2, origin, httpRewrites, httpsRewrites, pluginPath) {
   if (httpRewrites.length === 0 && httpsRewrites.length === 0 && !pluginPath) return void 0;
   let root = {};
@@ -59023,6 +59309,7 @@ async function runLaunch(params, deps = {}) {
   let piTmpHome;
   let ompTmpHome;
   let opencodeTmpFile;
+  let hermesTmpHome;
   const tmpFiles = [];
   const directUrl = launcherDirectUrl(process.env);
   if (directUrl) {
@@ -59055,6 +59342,16 @@ async function runLaunch(params, deps = {}) {
     const opencodePluginPath = opencodePlugin && fs10.existsSync(opencodePlugin) ? opencodePlugin : void 0;
     opencodeTmpFile = prepareOpencodeHttpRewrite(resolveOpencodeConfigFile(process.env), origin, routes.httpRewrites, routes.httpsRewrites, opencodePluginPath);
     if (opencodeTmpFile) env.OPENCODE_CONFIG = opencodeTmpFile;
+  } else if (base === "hermes") {
+    env = { ...process.env };
+    hermesTmpHome = prepareHermesHome(resolveHermesHome(process.env), origin, routes.httpRewrites);
+    if (hermesTmpHome) {
+      env.HERMES_HOME = hermesTmpHome;
+    } else {
+      console.error(
+        routes.httpRewrites.length === 0 ? "bili: no hermes providers found in ~/.hermes/config.yaml \u2014 traffic will NOT go through the proxy (configure a provider first)." : "bili: hermes config.yaml could not be rewritten (unreadable or no matching provider endpoints) \u2014 traffic will NOT go through the proxy."
+      );
+    }
   } else if (base === "codex") {
     const codexConversationId = injectMcp ? randomUUID2() : void 0;
     if (directUrl) {
@@ -59099,6 +59396,12 @@ async function runLaunch(params, deps = {}) {
       } catch {
       }
     }
+    if (hermesTmpHome) {
+      try {
+        fs10.rmSync(hermesTmpHome, { recursive: true, force: true });
+      } catch {
+      }
+    }
     if (opencodeTmpFile) {
       try {
         fs10.rmSync(path13.dirname(opencodeTmpFile), { recursive: true, force: true });
@@ -59333,6 +59636,7 @@ Usage:
   bili claude [opts --] [args]     start a proxy + launch claude against it (cert-MITM)
   bili omp [opts --] [args]        start a proxy + launch omp against it (cert-MITM)
   bili opencode [opts --] [args]   start a proxy + launch opencode against it (cert-MITM)
+  bili hermes [opts --] [args]     start a proxy + launch hermes-agent against it (/bili/ rewrite)
   bili test pi                     non-polluting pi smoke test through the proxy
   bili export [session] [--full]   list sessions / export one as a Markdown handoff
                                     (--full includes original messages; --output FILE)
@@ -59347,7 +59651,7 @@ Usage:
   bili --version                   print version
   bili --help                      show this help
 
-Launcher (bili pi / bili codex / bili claude / bili omp / bili opencode):
+Launcher (bili pi / bili codex / bili claude / bili omp / bili opencode / bili hermes):
   Brings up a proxy on an independent port (a fresh instance every launch), then runs the client pointed at it via HTTPS_PROXY + the proxy's
   MITM CA \u2014 no config-file edits. Discovered HTTPS upstream domains are
   auto-whitelisted for MITM so the proxy TLS-terminates exactly the hosts the
@@ -59359,6 +59663,7 @@ Launcher (bili pi / bili codex / bili claude / bili omp / bili opencode):
     bili codex                            # launch codex through the proxy
     bili claude                           # launch claude through the proxy
     bili omp                              # launch omp through the proxy (pi-based; /bili/ rewrite)
+    bili hermes                           # launch hermes-agent through the proxy (/bili/ rewrite of ~/.hermes/config.yaml)
     bili test pi                          # quick end-to-end check of the pi path
     bili pi --mitm-domain api.foo.com     # add a domain to the MITM whitelist