@code-yeongyu/senpi-ai 2026.10.15 → 2026.10.16-3

package/dist/api/google-shared.js

@@ -2,6 +2,7 @@
  * Shared utilities for Google Generative AI and Google Vertex providers.
  */
 import { FinishReason, FunctionCallingConfigMode, ThinkingLevel as GoogleSdkThinkingLevel, } from "@google/genai";
+import { isGoogleGeneration4OrLaterModelId } from "../google-thinking-generation.js";
 import { clampThinkingLevel } from "../models.js";
 import { retryProviderRequest } from "../utils/provider-retry.js";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
@@ -41,6 +42,8 @@ export function usesGoogleThinkingLevel(model) {
     // Match Gemini 3 Pro/Flash IDs with or without a minor version, such as
     // gemini-3-flash-preview, gemini-3.1-pro-preview, and gemini-3.8-flash.
     /gemini-3(?:\.\d+)?-(?:pro|flash)/.test(id) ||
+        // Generation 4 and later only accept discrete levels; a budget is rejected with HTTP 400 (senpi#3292).
+        isGoogleGeneration4OrLaterModelId(id) ||
         id === "gemini-flash-latest" ||
         id === "gemini-flash-lite-latest" ||
         // Match both hosted Gemma 4 naming forms: gemma-4-* and gemma4-*.

package/dist/api/google-vertex-adc.d.ts

@@ -0,0 +1,20 @@
+import type { ProviderEnv } from "../types.ts";
+export interface VertexAdcAccess {
+    accessToken: string;
+    /** Project whose Vertex endpoints the request targets. */
+    project: string;
+    /** Project billed for quota (`x-goog-user-project`); the ADC quota project when set. */
+    quotaProject: string;
+}
+/**
+ * Resolves an access token plus target and quota projects from Application
+ * Default Credentials, the same credential chain `@google/genai` uses for the
+ * google-vertex provider. Project precedence: explicit option, then
+ * GOOGLE_CLOUD_PROJECT/GCLOUD_PROJECT, then the ADC quota project, then the
+ * project google-auth-library discovers (gcloud config, metadata server).
+ */
+export declare function getVertexAdcAccess(options?: {
+    project?: string;
+    env?: ProviderEnv;
+}): Promise<VertexAdcAccess>;
+//# sourceMappingURL=google-vertex-adc.d.ts.map

package/dist/api/google-vertex-adc.js

@@ -0,0 +1,38 @@
+import { GoogleAuth } from "google-auth-library";
+import { buildGoogleAuthOptions, resolveVertexProjectFromEnv } from "./google-vertex-env.js";
+const CLOUD_PLATFORM_SCOPE = "https://www.googleapis.com/auth/cloud-platform";
+const authByKey = new Map();
+function getGoogleAuth(env) {
+    const authOptions = buildGoogleAuthOptions(env);
+    const key = authOptions?.keyFilename ?? "";
+    let auth = authByKey.get(key);
+    if (!auth) {
+        // One GoogleAuth per credential source so the library's token cache and refresh are reused.
+        auth = new GoogleAuth({ ...authOptions, scopes: [CLOUD_PLATFORM_SCOPE] });
+        authByKey.set(key, auth);
+    }
+    return auth;
+}
+/**
+ * Resolves an access token plus target and quota projects from Application
+ * Default Credentials, the same credential chain `@google/genai` uses for the
+ * google-vertex provider. Project precedence: explicit option, then
+ * GOOGLE_CLOUD_PROJECT/GCLOUD_PROJECT, then the ADC quota project, then the
+ * project google-auth-library discovers (gcloud config, metadata server).
+ */
+export async function getVertexAdcAccess(options) {
+    const auth = getGoogleAuth(options?.env);
+    const client = await auth.getClient();
+    const { token } = await client.getAccessToken();
+    if (!token)
+        throw new Error("Google Application Default Credentials returned no access token.");
+    const adcQuotaProject = client.quotaProjectId || undefined;
+    const project = resolveVertexProjectFromEnv(options) ??
+        adcQuotaProject ??
+        ((await auth.getProjectId().catch(() => null)) || undefined);
+    if (!project) {
+        throw new Error("Vertex AI requires a project ID. Set GOOGLE_CLOUD_PROJECT/GCLOUD_PROJECT or a quota project in Application Default Credentials.");
+    }
+    return { accessToken: token, project, quotaProject: adcQuotaProject ?? project };
+}
+//# sourceMappingURL=google-vertex-adc.js.map

package/dist/api/google-vertex-adc.lazy.d.ts

@@ -0,0 +1,9 @@
+import type { ProviderEnv } from "../types.ts";
+import type { VertexAdcAccess } from "./google-vertex-adc.ts";
+export type { VertexAdcAccess } from "./google-vertex-adc.ts";
+/** Lazily loads google-auth-library so importing callers do not pay for it until a request is made. */
+export declare function getVertexAdcAccess(options?: {
+    project?: string;
+    env?: ProviderEnv;
+}): Promise<VertexAdcAccess>;
+//# sourceMappingURL=google-vertex-adc.lazy.d.ts.map

package/dist/api/google-vertex-adc.lazy.js

@@ -0,0 +1,6 @@
+/** Lazily loads google-auth-library so importing callers do not pay for it until a request is made. */
+export async function getVertexAdcAccess(options) {
+    const module = await import("./google-vertex-adc.js");
+    return module.getVertexAdcAccess(options);
+}
+//# sourceMappingURL=google-vertex-adc.lazy.js.map

package/dist/api/google-vertex-env.d.ts

@@ -0,0 +1,9 @@
+import type { ProviderEnv } from "../types.ts";
+export declare function resolveVertexProjectFromEnv(options?: {
+    project?: string;
+    env?: ProviderEnv;
+}): string | undefined;
+export declare function buildGoogleAuthOptions(env?: ProviderEnv): {
+    keyFilename: string;
+} | undefined;
+//# sourceMappingURL=google-vertex-env.d.ts.map

package/dist/api/google-vertex-env.js

@@ -0,0 +1,11 @@
+import { getProviderEnvValue } from "../utils/provider-env.js";
+export function resolveVertexProjectFromEnv(options) {
+    return (options?.project ||
+        getProviderEnvValue("GOOGLE_CLOUD_PROJECT", options?.env) ||
+        getProviderEnvValue("GCLOUD_PROJECT", options?.env));
+}
+export function buildGoogleAuthOptions(env) {
+    const keyFilename = getProviderEnvValue("GOOGLE_APPLICATION_CREDENTIALS", env);
+    return keyFilename ? { keyFilename } : undefined;
+}
+//# sourceMappingURL=google-vertex-env.js.map

package/dist/api/google-vertex.js

@@ -9,6 +9,7 @@ import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
 import { getSystemMessageText } from "../utils/text.js";
 import { collapseSystemMessages, getCurrentTools, getInitialSystemMessage } from "../utils/transcript.js";
 import { convertMessages, convertTools, getDisabledGoogleThinkingConfig, isThinkingPart, mapStopReason, resolveGoogleFunctionCallingMode, resolveGoogleThinkingLevel, retainThoughtSignature, retryGoogleRequest, supportsGoogleStrictToolSampling, toGoogleSdkThinkingLevel, toGoogleThinkingLevel, toProviderNativeContent, usesGoogleThinkingLevel, } from "./google-shared.js";
+import { buildGoogleAuthOptions, resolveVertexProjectFromEnv } from "./google-vertex-env.js";
 import { applyExtraBody, buildBaseOptions, GOOGLE_RESERVED_BODY_KEYS } from "./simple-options.js";
 const API_VERSION = "v1";
 const GCP_VERTEX_CREDENTIALS_MARKER = "gcp-vertex-credentials";
@@ -351,10 +352,6 @@ function baseUrlIncludesApiVersion(baseUrl) {
         return /(?:^|\/)v\d+(?:beta\d*)?(?:\/|$)/.test(baseUrl);
     }
 }
-function buildGoogleAuthOptions(env) {
-    const keyFilename = getProviderEnvValue("GOOGLE_APPLICATION_CREDENTIALS", env);
-    return keyFilename ? { keyFilename } : undefined;
-}
 function resolveApiKey(options) {
     const apiKey = options?.apiKey?.trim();
     if (!apiKey || apiKey === GCP_VERTEX_CREDENTIALS_MARKER || isPlaceholderApiKey(apiKey)) {
@@ -366,9 +363,7 @@ function isPlaceholderApiKey(apiKey) {
     return /^<[^>]+>$/.test(apiKey);
 }
 function resolveProject(options) {
-    const project = options?.project ||
-        getProviderEnvValue("GOOGLE_CLOUD_PROJECT", options?.env) ||
-        getProviderEnvValue("GCLOUD_PROJECT", options?.env);
+    const project = resolveVertexProjectFromEnv(options);
     if (!project) {
         throw new Error("Vertex AI requires a project ID. Set GOOGLE_CLOUD_PROJECT/GCLOUD_PROJECT or pass project in options.");
     }

package/dist/api/openai-codex-responses/fallback-state.d.ts

@@ -1,4 +1,11 @@
 export declare const CODEX_WEBSOCKET_FALLBACK_COOLDOWN_MS = 60000;
+/**
+ * Mid-stream WebSocket failures in a row, with no completed WebSocket response
+ * between them, before the session moves to SSE. One drop is transient: the
+ * retry opens a fresh WebSocket, as upstream Codex does (senpi#3442).
+ */
+export declare const CODEX_WEBSOCKET_MID_STREAM_FAILURES_BEFORE_FALLBACK = 2;
+export type WebSocketFailurePhase = "before_message_stream_start" | "after_message_stream_start";
 export interface ChatGptSubscriptionWebSocketDebugStats {
     requests: number;
     connectionsCreated: number;
@@ -20,5 +27,14 @@ export declare function getWebSocketDebugStats(sessionId: string): ChatGptSubscr
 export declare function clearWebSocketFallbackState(sessionId?: string): void;
 export declare function isWebSocketSseFallbackActive(sessionId: string | undefined): boolean;
 export declare function recordWebSocketSseFallback(sessionId: string | undefined): void;
-export declare function recordWebSocketFailure(sessionId: string | undefined, error: unknown): void;
+/**
+ * A failure before any output arms the SSE cooldown at once, because that same
+ * request is retried over SSE. A failure after output started is retried by the
+ * caller; it arms the cooldown only once it repeats without a completed
+ * WebSocket response in between. Aborts and non-transport errors never reach
+ * here, so they neither count nor reset the run.
+ */
+export declare function recordWebSocketFailure(sessionId: string | undefined, error: unknown, phase: WebSocketFailurePhase): void;
+/** A completed WebSocket response ends a run of mid-stream failures. */
+export declare function recordWebSocketSuccess(sessionId: string | undefined): void;
 //# sourceMappingURL=fallback-state.d.ts.map

package/dist/api/openai-codex-responses/fallback-state.js

@@ -1,7 +1,14 @@
 import { formatThrownValue } from "../../utils/diagnostics.js";
 export const CODEX_WEBSOCKET_FALLBACK_COOLDOWN_MS = 60_000;
+/**
+ * Mid-stream WebSocket failures in a row, with no completed WebSocket response
+ * between them, before the session moves to SSE. One drop is transient: the
+ * retry opens a fresh WebSocket, as upstream Codex does (senpi#3442).
+ */
+export const CODEX_WEBSOCKET_MID_STREAM_FAILURES_BEFORE_FALLBACK = 2;
 const websocketDebugStats = new Map();
 const websocketSseFallbackUntil = new Map();
+const consecutiveMidStreamFailures = new Map();
 export function getOrCreateWebSocketDebugStats(sessionId) {
     let stats = websocketDebugStats.get(sessionId);
     if (!stats) {
@@ -29,10 +36,12 @@ export function clearWebSocketFallbackState(sessionId) {
     if (sessionId) {
         websocketDebugStats.delete(sessionId);
         websocketSseFallbackUntil.delete(sessionId);
+        consecutiveMidStreamFailures.delete(sessionId);
         return;
     }
     websocketDebugStats.clear();
     websocketSseFallbackUntil.clear();
+    consecutiveMidStreamFailures.clear();
 }
 export function isWebSocketSseFallbackActive(sessionId) {
     if (!sessionId)
@@ -58,13 +67,33 @@ export function recordWebSocketSseFallback(sessionId) {
     stats.sseFallbacks++;
     stats.websocketFallbackActive = true;
 }
-export function recordWebSocketFailure(sessionId, error) {
+/**
+ * A failure before any output arms the SSE cooldown at once, because that same
+ * request is retried over SSE. A failure after output started is retried by the
+ * caller; it arms the cooldown only once it repeats without a completed
+ * WebSocket response in between. Aborts and non-transport errors never reach
+ * here, so they neither count nor reset the run.
+ */
+export function recordWebSocketFailure(sessionId, error, phase) {
     if (!sessionId)
         return;
-    websocketSseFallbackUntil.set(sessionId, Date.now() + CODEX_WEBSOCKET_FALLBACK_COOLDOWN_MS);
     const stats = getOrCreateWebSocketDebugStats(sessionId);
     stats.websocketFailures++;
     stats.lastWebSocketError = formatThrownValue(error);
+    if (phase === "after_message_stream_start") {
+        const failures = (consecutiveMidStreamFailures.get(sessionId) ?? 0) + 1;
+        if (failures < CODEX_WEBSOCKET_MID_STREAM_FAILURES_BEFORE_FALLBACK) {
+            consecutiveMidStreamFailures.set(sessionId, failures);
+            return;
+        }
+    }
+    consecutiveMidStreamFailures.delete(sessionId);
+    websocketSseFallbackUntil.set(sessionId, Date.now() + CODEX_WEBSOCKET_FALLBACK_COOLDOWN_MS);
     stats.websocketFallbackActive = true;
 }
+/** A completed WebSocket response ends a run of mid-stream failures. */
+export function recordWebSocketSuccess(sessionId) {
+    if (sessionId)
+        consecutiveMidStreamFailures.delete(sessionId);
+}
 //# sourceMappingURL=fallback-state.js.map

package/dist/api/openai-codex-responses.js

@@ -21,7 +21,7 @@ import { getSystemMessageText } from "../utils/text.js";
 import { getDeclaredTools, getInitialSystemMessage, normalizeContext, resolveTranscript } from "../utils/transcript.js";
 import { uuidv7 } from "../utils/uuid.js";
 import { createGrammarToolInputProperties } from "./constrained-sampling.js";
-import { clearWebSocketFallbackState, getOrCreateWebSocketDebugStats, getWebSocketDebugStats, isWebSocketSseFallbackActive, recordWebSocketFailure, recordWebSocketSseFallback, } from "./openai-codex-responses/fallback-state.js";
+import { clearWebSocketFallbackState, getOrCreateWebSocketDebugStats, getWebSocketDebugStats, isWebSocketSseFallbackActive, recordWebSocketFailure, recordWebSocketSseFallback, recordWebSocketSuccess, } from "./openai-codex-responses/fallback-state.js";
 import { buildCodexReasoning } from "./openai-codex-responses/reasoning.js";
 import { applyChatGptSubscriptionCacheAffinityHeaders, clampOpenAIPromptCacheKey } from "./openai-prompt-cache.js";
 import { convertResponsesMessages, convertResponsesTools, processResponsesStream, resolveResponsesDeferredToolsMode, resolveResponsesToolPlacement, } from "./openai-responses-shared.js";
@@ -202,6 +202,7 @@ export const stream = (model, context, options) => {
                             throw new Error("Request was aborted");
                         }
                         assertSuccessfulOutput(output);
+                        recordWebSocketSuccess(cacheSessionId);
                         stream.push({
                             type: "done",
                             reason: output.stopReason,
@@ -225,14 +226,15 @@ export const stream = (model, context, options) => {
                         if (aborted || (isCodexNonTransportError(error) && !connectionLimitBeforeStart)) {
                             throw error;
                         }
+                        const phase = websocketStarted ? "after_message_stream_start" : "before_message_stream_start";
                         appendAssistantMessageDiagnostic(output, createAssistantMessageDiagnostic("provider_transport_failure", error, {
                             configuredTransport: transport,
                             ...(websocketStarted ? {} : { fallbackTransport: "sse" }),
                             eventsEmitted: websocketStarted,
-                            phase: websocketStarted ? "after_message_stream_start" : "before_message_stream_start",
+                            phase,
                             requestBytes: new TextEncoder().encode(bodyJson).byteLength,
                         }));
-                        recordWebSocketFailure(cacheSessionId, error);
+                        recordWebSocketFailure(cacheSessionId, error, phase);
                         if (websocketStarted) {
                             throw error;
                         }

package/dist/google-thinking-generation.d.ts

@@ -0,0 +1,5 @@
+import type { Api, Model, ThinkingLevelMap } from "./types.ts";
+export declare function isGoogleGeneration4OrLaterModelId(modelId: string): boolean;
+/** The level map for a generation-4-or-later row that ships none; generated metadata always wins. */
+export declare function inferGoogleThinkingLevelMap<TApi extends Api>(model: Model<TApi>): ThinkingLevelMap | undefined;
+//# sourceMappingURL=google-thinking-generation.d.ts.map

package/dist/google-thinking-generation.js

@@ -0,0 +1,29 @@
+const GOOGLE_THINKING_APIS = ["google-generative-ai", "google-vertex"];
+/**
+ * Generation 4 and later of this provider's models accept only the discrete low, medium and high
+ * levels: a token budget, `minimal`, or turning thinking off is rejected with HTTP 400 (senpi#3292).
+ */
+const GENERATION_4_THINKING_LEVEL_MAP = {
+    off: null,
+    minimal: null,
+    low: "low",
+    medium: "medium",
+    high: "high",
+    xhigh: null,
+    max: null,
+};
+/** A complete id or a `models/`-style namespace, generation 4 or later, then the end or a `-` suffix. */
+const GENERATION_4_ID = /(?:^|\/)gemini-(?:[4-9]|[1-9]\d)(?:\.\d+)?(?:-|$)/;
+export function isGoogleGeneration4OrLaterModelId(modelId) {
+    return GENERATION_4_ID.test(modelId.toLowerCase());
+}
+/** The level map for a generation-4-or-later row that ships none; generated metadata always wins. */
+export function inferGoogleThinkingLevelMap(model) {
+    if (model.thinkingLevelMap !== undefined)
+        return model.thinkingLevelMap;
+    if (GOOGLE_THINKING_APIS.includes(model.api) && isGoogleGeneration4OrLaterModelId(model.id)) {
+        return GENERATION_4_THINKING_LEVEL_MAP;
+    }
+    return undefined;
+}
+//# sourceMappingURL=google-thinking-generation.js.map

package/dist/index.d.ts

@@ -34,6 +34,7 @@ export * from "./endpoint-reasoning-efforts.ts";
 export * from "./env-api-keys.ts";
 export * from "./legacy-provider-ids.ts";
 export { modelSupportsAssistantPrefill } from "./model.ts";
+export * from "./model-cost.ts";
 export * from "./models.ts";
 export * from "./models-store.ts";
 export { supportsAllowedToolChoice } from "./openai-responses-compat.ts";

package/dist/index.js

@@ -20,6 +20,7 @@ export * from "./endpoint-reasoning-efforts.js";
 export * from "./env-api-keys.js";
 export * from "./legacy-provider-ids.js";
 export { modelSupportsAssistantPrefill } from "./model.js";
+export * from "./model-cost.js";
 export * from "./models.js";
 export * from "./models-store.js";
 export { supportsAllowedToolChoice } from "./openai-responses-compat.js";

package/dist/model-cost.d.ts

@@ -0,0 +1,4 @@
+import type { ModelCost, ModelCostCard } from "./types.ts";
+/** The card in force at `atMs`: the latest dated card that has started, else the base card. */
+export declare function costCardAt(cost: ModelCost, atMs: number): ModelCostCard;
+//# sourceMappingURL=model-cost.d.ts.map

package/dist/model-cost.js

@@ -0,0 +1,14 @@
+/** The card in force at `atMs`: the latest dated card that has started, else the base card. */
+export function costCardAt(cost, atMs) {
+    let card = cost;
+    let startedAt = Number.NEGATIVE_INFINITY;
+    for (const rate of cost.effectiveRates ?? []) {
+        const from = Date.parse(rate.effectiveFrom);
+        if (from <= atMs && from > startedAt) {
+            card = rate;
+            startedAt = from;
+        }
+    }
+    return card;
+}
+//# sourceMappingURL=model-cost.js.map

package/dist/models.d.ts

@@ -272,9 +272,11 @@ export declare function createProvider<TApi extends Api = Api>(input: CreateProv
  * Non-chat models never match, even when their api id equals `api`.
  */
 export declare function hasApi<TApi extends Api>(model: AnyModel, api: TApi): model is Model<TApi>;
-export declare function calculateCost(model: AnyModel, usage: Usage): Usage["cost"];
+export declare function calculateCost(model: AnyModel, usage: Usage, atMs?: number): Usage["cost"];
 /** Infer documented OpenAI reasoning controls only when generated metadata is absent. */
 export declare function inferOpenAIThinkingLevelMap<TApi extends Api>(model: Model<TApi>): ThinkingLevelMap | undefined;
+/** The level map a row carries, else the one inferred from documented provider controls. */
+export declare function inferThinkingLevelMap<TApi extends Api>(model: Model<TApi>): ThinkingLevelMap | undefined;
 export declare function getSupportedThinkingLevels<TApi extends Api>(model: Model<TApi>): ModelThinkingLevel[];
 export declare function clampThinkingLevel<TApi extends Api>(model: Model<TApi>, level: ModelThinkingLevel): ModelThinkingLevel;
 /**

package/dist/models.js

@@ -4,7 +4,9 @@ import { InMemoryCredentialStore } from "./auth/credential-store.js";
 import { appendLoginSlot, removeSlot } from "./auth/pool/slots.js";
 import { resolveRefreshCredential } from "./auth/refresh-credential.js";
 import { ModelsError, providerNotConfiguredMessage, resolveProviderAuth, } from "./auth/resolve.js";
+import { inferGoogleThinkingLevelMap } from "./google-thinking-generation.js";
 import { builtinCatalogAdvertisesMax } from "./model-catalog.js";
+import { costCardAt } from "./model-cost.js";
 import { InMemoryModelsStore } from "./models-store.js";
 import { operationSignal, raceWithAbortSignal } from "./utils/abort.js";
 import { assertChatModel, assertClassifierModel, assertImageModel, classifierErrorResult, getModelType, imageErrorResult, isModelType, } from "./utils/model-operations.js";
@@ -688,11 +690,12 @@ export function createProvider(input) {
 export function hasApi(model, api) {
     return isModelType(model, "chat") && model.api === api;
 }
-export function calculateCost(model, usage) {
+export function calculateCost(model, usage, atMs = Date.now()) {
     const inputTokens = usage.input + usage.cacheRead + usage.cacheWrite;
-    let rates = model.cost;
+    const card = costCardAt(model.cost, atMs);
+    let rates = card;
     let matchedThreshold = -1;
-    for (const tier of model.cost.tiers ?? []) {
+    for (const tier of card.tiers ?? []) {
         if (inputTokens > tier.inputTokensAbove && tier.inputTokensAbove > matchedThreshold) {
             rates = tier;
             matchedThreshold = tier.inputTokensAbove;
@@ -736,11 +739,15 @@ export function inferOpenAIThinkingLevelMap(model) {
     }
     return undefined;
 }
+/** The level map a row carries, else the one inferred from documented provider controls. */
+export function inferThinkingLevelMap(model) {
+    return inferOpenAIThinkingLevelMap(model) ?? inferGoogleThinkingLevelMap(model);
+}
 export function getSupportedThinkingLevels(model) {
     if (!model.reasoning)
         return ["off"];
     return EXTENDED_THINKING_LEVELS.filter((level) => {
-        const mapped = inferOpenAIThinkingLevelMap(model)?.[level];
+        const mapped = inferThinkingLevelMap(model)?.[level];
         if (mapped === null)
             return false;
         if (level === "xhigh")

package/dist/provider-diagnostic.d.ts

@@ -1,8 +1,8 @@
 import { PROVIDER_DIAGNOSTIC_MAX_BYTES } from "./utils/provider-diagnostic-vocabulary.ts";
 /** Closed failure families. `unknown` is kept rather than guessed. */
-export type ProviderDiagnosticCategory = "auth" | "rate_limit" | "quota" | "context_limit" | "invalid_request" | "provider_unavailable" | "unknown";
+export type ProviderDiagnosticCategory = "auth" | "rate_limit" | "quota" | "context_limit" | "invalid_request" | "provider_unavailable" | "network" | "unknown";
 /** Which structured provider field decided the category. */
-export type ProviderDiagnosticEvidence = "structured_status" | "structured_code";
+export type ProviderDiagnosticEvidence = "structured_status" | "structured_code" | "structured_transport";
 /**
  * Bounded, redaction-safe classification of a provider rejection, minted only by
  * a provider adapter from structured metadata (the SDK HTTP error its own

package/dist/provider-diagnostic.js

@@ -1,5 +1,5 @@
 import { peekProviderDiagnostic } from "./utils/provider-diagnostic-carrier.js";
-import { buildProviderDiagnostic, categoryForProviderCode, categoryForProviderStatus, isProviderCodeStatusCompatible, isProviderDiagnosticCategory, isProviderDiagnosticStatus, PROVIDER_DIAGNOSTIC_MAX_BYTES, } from "./utils/provider-diagnostic-vocabulary.js";
+import { buildProviderDiagnostic, categoryForProviderCode, categoryForProviderStatus, categoryForTransportCode, isProviderCodeStatusCompatible, isProviderDiagnosticCategory, isProviderDiagnosticStatus, PROVIDER_DIAGNOSTIC_MAX_BYTES, } from "./utils/provider-diagnostic-vocabulary.js";
 export { PROVIDER_DIAGNOSTIC_MAX_BYTES };
 function readField(value, key) {
     return Reflect.get(value, key);
@@ -28,6 +28,13 @@ export function sanitizeProviderDiagnostic(value) {
                 return undefined;
             return buildProviderDiagnostic({ category, httpStatus, evidence });
         }
+        if (evidence === "structured_transport") {
+            if (httpStatus !== undefined || typeof code !== "string")
+                return undefined;
+            if (categoryForTransportCode(code) !== category)
+                return undefined;
+            return buildProviderDiagnostic({ category, code, evidence });
+        }
         if (evidence !== "structured_code" || typeof code !== "string")
             return undefined;
         if (categoryForProviderCode(code) !== category)

package/dist/providers/all.js

@@ -69,10 +69,12 @@ const XIAOMI_MIMO_PROVIDERS = new Set([
     "xiaomi-token-plan-ams",
     "xiaomi-token-plan-sgp",
 ]);
+// mimo-v2.5-pro is deprecated upstream (models.dev) and drops out of a regenerated catalog; v2.6-pro is its successor.
+const XIAOMI_MIMO_PRO_IDS = new Set(["mimo-v2.5-pro", "mimo-v2.6-pro"]);
 function normalizeBuiltinModel(model) {
     if (!model)
         return undefined;
-    if (XIAOMI_MIMO_PROVIDERS.has(model.provider) && model.id === "mimo-v2.5-pro") {
+    if (XIAOMI_MIMO_PROVIDERS.has(model.provider) && XIAOMI_MIMO_PRO_IDS.has(model.id)) {
         return {
             ...model,
             compat: {

package/dist/providers/data/.manifest.json

@@ -1 +1 @@
-{"schemaVersion":6,"generatedAt":"2026-10-15T02:44:52.929Z","structureHash":"9d1dc6dea8552f60e610d04360a6d720f331b7b963dbe742cb6f4a2f73938910","files":{"alibaba-token-plan.json":"16b3c6c78d72c73ca34a1e79ce71c0cf36ba409d02e83a17396ce0949a1d5363","amazon-bedrock.json":"4a95e6bf974e5ca967d5c050df580262e13573e42b3186edb448fa11a302f09f","ant-ling.json":"628feab3ef6c7d80ce96f75999804e1d9f05fec3851e5302e30fff3b69db90f6","anthropic.json":"111384e9f5845e0247bd34edf2400ef7720fcbdac140b944cf8878192d8bd6ab","azure-openai-responses.json":"a3e52bd78f0220aa32bce519e13955b8e6e567860416343844cbe19902f4a682","bai.json":"7e6861aaf27d8af34e1771c6092602194fcec0a0046e6adfab9c3ee54d32b99f","baseten.json":"1fcdab8daf2881bd099eb08f9416f7df5bf39975937d2a63b841542bd032f2e6","cerebras.json":"3eccb58c5de3d0dc7ba49dc6e477ba29f00cb3150ce0a35cc90d59e6b5ab368e","chatgpt-subscription.json":"42b88c93d717b2a7469b71da7da5adf87347d0d9cc0f264b39117f0f940b51c0","cloudflare-ai-gateway.json":"673ac36e4531995c43e49e132577cdde18a0ba088648ecae5a904020f400abc2","cloudflare-workers-ai.json":"f351e80493dea7b7a2c31dceb086ee804a88d67c5f8a3c823ea907991fe5d659","deepseek.json":"10a296fb3e898f7715c80890c8af0af4f5fd58f22dbcd6612fdc5d762157ccdc","fireworks.json":"c3dc21dc2e95428fc9a38e3840c05bbf90f94e4c25dee0eedf0b385b49ff65a4","github-copilot.json":"9a0a960b5978134bdd7a5e56928fb1985a10ce893fd31d93a272a4b38c413740","google-vertex.json":"564d692a4292ef12e98c8ec056a39948ea64e10851f5ddaaebcc37b585545231","google.json":"a8fd7cc81346d7afd5f58df3e5b91012ba75721dfae99c31c69c623d505ca1b0","groq.json":"910952ad9dad07856983459d63fce0148dba1203edd7b10d801a77c5824f3405","huggingface.json":"6399e958790a9d90c21fb55d1cbef8569427a1ceb28de86dadeb3b4454ecfade","meta.json":"563f5ec07973f52ab8c3eb26ef91a36ad8a79e950b37174935666cdcb624bf35","minimax-cn.json":"f962c20ee6a6a342820f778e9dc363a18a524bf11998f6e2c57f2989915d62e5","minimax.json":"17a432f80d0ea6dd26eade80a4314250d3a855678601728a5f70dd8fa0d21dd2","mistral.json":"27aa66c71a0504b53dc87ec194286fa7e70b99747fc8df9b709b23b3548afe50","moonshotai-cn.json":"18954b55dc5edecc174d8a07a2c5993416dd9d3ce52e411438035c16ee982e8e","moonshotai.json":"62d08fd5beead025a56d5f2e4227d8516b9c5034275b1b36d761a8625918f2b9","nvidia.json":"cc7b418f706178f0aac2384324cd63f7641ff03656ef25551a1859e2c69381bc","openai.json":"9104538118053eed57f90852332b6ad274a4e4792389bdbc9cfca1ae400c705a","opencode-go.json":"aa6e912d5aa4cb3f810f0c90a9a8ff033cab51f2e96d0daff0bcd0d2e5c4816a","opencode.json":"9d8b40241035e04a86bbcadd737308701b797630e1ed5c10ac28b2005991f01f","opengateway.json":"0e6b14bcc26e49943f10fc9400a895759bb644254fbf201115f6d6a345615bab","openrouter.json":"1cb71cae6b9e9511268b0a87079fb84bd6b306af0d8149825f9bdaa4e623ee38","qwen-token-plan-cn.json":"bd36a6d9cc2048deafacc85f7f0f2e51868c8b7aa09502e3e4e60e3e8b76f690","qwen-token-plan-individual.json":"35ffb0ba1a1ce3f68fc73262dd5e07fbcb5bf5911ad1e6e17bc0621e2894676a","qwen-token-plan.json":"ca27e49e62965221badead8e867b2007035c6b6a15a97c47219bd23d2c83c2b1","radius.json":"e33fab6aa5ac6c43ce26dc3680ea28f2d7d063a8dbca0360ee2b9c5676bc43a6","together.json":"bbd7fad86cacfefe3d3d61f6638f22e9b6f5d575722fd1112a16aafe0485c46d","typesafe.json":"a4429a2f696d5cc5273969af61a7310bda07d6bdcd27bf9ac9c76c534cc43716","venice.json":"8ca5cd7b4fba5aec009b53607daa0929de9b9bc62008779383a16332433afcd1","vercel-ai-gateway.json":"2deb62b526f3bc3369dc850cc3d7e2f365293b75ebd55f1e5c846c0fa89e765f","xai.json":"d7771a1450b21c1f6aee887ca6fea61b6e7b3fc13b4dee5cf72a71134fe623bc","xiaomi-token-plan-ams.json":"4dd8f838e21d7aa7caf3122d0b0b1e5a02a33bc012b96e6a75bd119334152d7a","xiaomi-token-plan-cn.json":"46414ff49973b95a0c14bd33e8992cc1938a50ce2c7cea9e2e5cc02de8f1df31","xiaomi-token-plan-sgp.json":"f5b3f66bf7bb7238592d4a9667cad983a008d7539a66e9896b02422622800935","xiaomi.json":"8b62c158d55d5a165f35cd72c1fc82ab89ff8e73f08960105fc41b822f236f69","zai-coding-cn.json":"2a7f59d15f588a97d218997883f2506dfc52d3ee320697d39f14213a050fc912","zai.json":"e9cc408e1756de8abc70148d63d89d11a63c65b7ef06ad0661cd1328e4209ba3"}}
+{"schemaVersion":6,"generatedAt":"2026-10-16T08:15:38.355Z","structureHash":"4b14b62036b26c05fb7e26d1b722135a8249b002d357c71bda326cfef4b2021e","files":{"alibaba-token-plan.json":"16b3c6c78d72c73ca34a1e79ce71c0cf36ba409d02e83a17396ce0949a1d5363","amazon-bedrock.json":"4a95e6bf974e5ca967d5c050df580262e13573e42b3186edb448fa11a302f09f","ant-ling.json":"628feab3ef6c7d80ce96f75999804e1d9f05fec3851e5302e30fff3b69db90f6","anthropic.json":"111384e9f5845e0247bd34edf2400ef7720fcbdac140b944cf8878192d8bd6ab","azure-openai-responses.json":"a3e52bd78f0220aa32bce519e13955b8e6e567860416343844cbe19902f4a682","bai.json":"7e6861aaf27d8af34e1771c6092602194fcec0a0046e6adfab9c3ee54d32b99f","baseten.json":"1fcdab8daf2881bd099eb08f9416f7df5bf39975937d2a63b841542bd032f2e6","cerebras.json":"3eccb58c5de3d0dc7ba49dc6e477ba29f00cb3150ce0a35cc90d59e6b5ab368e","chatgpt-subscription.json":"42b88c93d717b2a7469b71da7da5adf87347d0d9cc0f264b39117f0f940b51c0","cloudflare-ai-gateway.json":"673ac36e4531995c43e49e132577cdde18a0ba088648ecae5a904020f400abc2","cloudflare-workers-ai.json":"f351e80493dea7b7a2c31dceb086ee804a88d67c5f8a3c823ea907991fe5d659","deepseek.json":"10a296fb3e898f7715c80890c8af0af4f5fd58f22dbcd6612fdc5d762157ccdc","fireworks.json":"868f828365314089608e6d0d915076f6d3d4f67c305f824d47ba05e040897d39","github-copilot.json":"9a0a960b5978134bdd7a5e56928fb1985a10ce893fd31d93a272a4b38c413740","google-vertex.json":"8f986b4aca8c685b0f3481bd252bda21e4858bf9015ad37fc5bcc44b7f01d076","google.json":"fc06c3cd56c180d214612ce0ef72ea14734195de70ace47ef47f02199b053aae","groq.json":"910952ad9dad07856983459d63fce0148dba1203edd7b10d801a77c5824f3405","huggingface.json":"6399e958790a9d90c21fb55d1cbef8569427a1ceb28de86dadeb3b4454ecfade","meta.json":"563f5ec07973f52ab8c3eb26ef91a36ad8a79e950b37174935666cdcb624bf35","minimax-cn.json":"f962c20ee6a6a342820f778e9dc363a18a524bf11998f6e2c57f2989915d62e5","minimax.json":"17a432f80d0ea6dd26eade80a4314250d3a855678601728a5f70dd8fa0d21dd2","mistral.json":"27aa66c71a0504b53dc87ec194286fa7e70b99747fc8df9b709b23b3548afe50","moonshotai-cn.json":"18954b55dc5edecc174d8a07a2c5993416dd9d3ce52e411438035c16ee982e8e","moonshotai.json":"62d08fd5beead025a56d5f2e4227d8516b9c5034275b1b36d761a8625918f2b9","nvidia.json":"98b47c4da26c8e649f48a874602fd7e9dc8a54ee0bdb40e3d07fe49189a842d2","openai.json":"9104538118053eed57f90852332b6ad274a4e4792389bdbc9cfca1ae400c705a","opencode-go.json":"b78a595ce56b3da17d93d609e490eda7843e090cc7943f96423c108ca2abbbc9","opencode.json":"bba280a58b195539d45ebff7583b0211b8c4326bfae3e3cf5cc17dd1dde9dae6","opengateway.json":"41e04d37ca867d533349c4dd37c9cf7274f6321194a57bbaa5ef2386b0cb0bf0","openrouter.json":"54abb027f2185f5f9b084aef4da30f0483bcfa894f68a934d588f144baea85c8","qwen-token-plan-cn.json":"bd36a6d9cc2048deafacc85f7f0f2e51868c8b7aa09502e3e4e60e3e8b76f690","qwen-token-plan-individual.json":"35ffb0ba1a1ce3f68fc73262dd5e07fbcb5bf5911ad1e6e17bc0621e2894676a","qwen-token-plan.json":"ca27e49e62965221badead8e867b2007035c6b6a15a97c47219bd23d2c83c2b1","radius.json":"e33fab6aa5ac6c43ce26dc3680ea28f2d7d063a8dbca0360ee2b9c5676bc43a6","together.json":"bbd7fad86cacfefe3d3d61f6638f22e9b6f5d575722fd1112a16aafe0485c46d","typesafe.json":"a4429a2f696d5cc5273969af61a7310bda07d6bdcd27bf9ac9c76c534cc43716","venice.json":"7ccc4284bb1bdd458898bc64dc0c2f07b629dcfc6cda41b13d1dc35feba98190","vercel-ai-gateway.json":"aff3af2ac9f49338f733825f0e801acb96ab0196062dccedd543ad86ef702c7b","xai.json":"d7771a1450b21c1f6aee887ca6fea61b6e7b3fc13b4dee5cf72a71134fe623bc","xiaomi-token-plan-ams.json":"7fa339ba5675460b1ff4e5bca9e8ad0aa60b0277d75a989b1c196aa6ca2af34d","xiaomi-token-plan-cn.json":"ef7917c6ea0d3edc655c8a52082ee5f1330658aa6f10199d8f164eb4605ef0a0","xiaomi-token-plan-sgp.json":"addda93f6b70ba879db97e3cfd2e4f2efc7cda47d7daa78c23c5b0e51162de5c","xiaomi.json":"5a42caf49d3e4d7946d60386e9899c2146bf8e3cbaff4c576ee4f49b398cc11c","zai-coding-cn.json":"2a7f59d15f588a97d218997883f2506dfc52d3ee320697d39f14213a050fc912","zai.json":"e9cc408e1756de8abc70148d63d89d11a63c65b7ef06ad0661cd1328e4209ba3"}}

package/dist/providers/data/fireworks.json

@@ -1 +1 @@
-{"anthropic-messages":{"chat:accounts/fireworks/models/deepseek-v4p1-flash":{"id":"accounts/fireworks/models/deepseek-v4p1-flash","name":"DeepSeek V4.1 Flash","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":0.3,"output":1.2,"cacheRead":0.006,"cacheWrite":0},"contextWindow":1000000,"maxTokens":384000,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/models/ember-1":{"id":"accounts/fireworks/models/ember-1","name":"Ember-1","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":3,"output":15,"cacheRead":0.3,"cacheWrite":0},"contextWindow":1048576,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":"medium","high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/models/gpt-oss-120b":{"id":"accounts/fireworks/models/gpt-oss-120b","name":"GPT OSS 120B","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":0.15,"output":0.6,"cacheRead":0.015,"cacheWrite":0},"contextWindow":131072,"maxTokens":32768,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":"medium","high":"high","xhigh":null,"max":null},"type":"chat"},"chat:accounts/fireworks/models/inkling":{"id":"accounts/fireworks/models/inkling","name":"Inkling","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":1,"output":4.05,"cacheRead":0.17,"cacheWrite":0},"contextWindow":1048576,"maxTokens":1048576,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/models/minimax-m3":{"id":"accounts/fireworks/models/minimax-m3","name":"MiniMax-M3","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":0.3,"output":1.2,"cacheRead":0.06,"cacheWrite":0},"contextWindow":512000,"maxTokens":512000,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":"medium","high":"high","xhigh":null,"max":null},"type":"chat"},"chat:accounts/fireworks/models/nemotron-3-ultra-nvfp4":{"id":"accounts/fireworks/models/nemotron-3-ultra-nvfp4","name":"Nemotron 3 Ultra 550B A55B","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":0.6,"output":2.4,"cacheRead":0.12,"cacheWrite":0},"contextWindow":262144,"maxTokens":128000,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},"type":"chat"},"chat:accounts/fireworks/models/nemotron-lightning-3p5-30b-a3b":{"id":"accounts/fireworks/models/nemotron-lightning-3p5-30b-a3b","name":"Nemotron 3.5 Lightning 30B A3B","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":0.05,"output":0.2,"cacheRead":0.01,"cacheWrite":0},"contextWindow":262144,"maxTokens":262144,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},"type":"chat"},"chat:accounts/fireworks/models/qwen3p8-2p4t-a95b":{"id":"accounts/fireworks/models/qwen3p8-2p4t-a95b","name":"Qwen3.8 2.4T A95B","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":2,"output":6,"cacheRead":0.25,"cacheWrite":0},"contextWindow":262144,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":"medium","high":null,"xhigh":"xhigh","max":null},"type":"chat"},"chat:accounts/fireworks/models/qwen3p8-max":{"id":"accounts/fireworks/models/qwen3p8-max","name":"Qwen3.8 Max","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":2,"output":6,"cacheRead":0.25,"cacheWrite":0},"contextWindow":262144,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":"medium","high":null,"xhigh":"xhigh","max":null},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/routers/deepseek-flash-latest":{"id":"accounts/fireworks/routers/deepseek-flash-latest","name":"DeepSeek Flash Latest","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":0.3,"output":1.2,"cacheRead":0.006,"cacheWrite":0},"contextWindow":1000000,"maxTokens":384000,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/routers/kimi-fast-latest":{"id":"accounts/fireworks/routers/kimi-fast-latest","name":"Kimi Fast Latest","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":4.5,"output":22.5,"cacheRead":0.45,"cacheWrite":0},"contextWindow":1048576,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":"medium","high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/routers/kimi-latest":{"id":"accounts/fireworks/routers/kimi-latest","name":"Kimi Latest","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":3,"output":15,"cacheRead":0.3,"cacheWrite":0},"contextWindow":1048576,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":"medium","high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/routers/minimax-latest":{"id":"accounts/fireworks/routers/minimax-latest","name":"MiniMax Latest","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":0.3,"output":1.2,"cacheRead":0.06,"cacheWrite":0},"contextWindow":512000,"maxTokens":512000,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":"medium","high":"high","xhigh":null,"max":null},"type":"chat"},"chat:accounts/fireworks/routers/qwen-max-latest":{"id":"accounts/fireworks/routers/qwen-max-latest","name":"Qwen Max Latest (Qwen3.8 Max)","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":2,"output":6,"cacheRead":0.25,"cacheWrite":0},"contextWindow":262144,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"}},"openai-completions":{"chat:accounts/fireworks/models/glm-5p3":{"id":"accounts/fireworks/models/glm-5p3","name":"GLM 5.3","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":1.4,"output":4.4,"cacheRead":0.26,"cacheWrite":0},"contextWindow":1048573,"maxTokens":262144,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false},"thinkingLevelMap":{"off":"none","minimal":null,"low":null,"medium":null,"high":"high","xhigh":null,"max":"max"},"type":"chat"},"chat:accounts/fireworks/models/glm-5p3-flash":{"id":"accounts/fireworks/models/glm-5p3-flash","name":"GLM 5.3 Flash","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":0.15,"output":0.5,"cacheRead":0.03,"cacheWrite":0},"contextWindow":1048573,"maxTokens":131072,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false},"thinkingLevelMap":{"off":"none","minimal":null,"low":null,"medium":null,"high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/models/kimi-k3":{"id":"accounts/fireworks/models/kimi-k3","name":"Kimi K3","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":3,"output":15,"cacheRead":0.3,"cacheWrite":0},"contextWindow":1048576,"maxTokens":131072,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false,"requiresReasoningContentOnAssistantMessages":true,"thinkingFormat":"openai","supportsMidConvoSystemMessages":true,"supportsMidConvoToolAdditions":true},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/routers/glm-5p3-fast":{"id":"accounts/fireworks/routers/glm-5p3-fast","name":"GLM 5.3 Fast","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":2.1,"output":6.6,"cacheRead":0.39,"cacheWrite":0},"contextWindow":1048572,"maxTokens":262144,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false},"thinkingLevelMap":{"off":"none","minimal":null,"low":null,"medium":null,"high":"high","xhigh":null,"max":"max"},"type":"chat"},"chat:accounts/fireworks/routers/glm-fast-latest":{"id":"accounts/fireworks/routers/glm-fast-latest","name":"GLM 5.3 Fast (Latest)","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":2.1,"output":6.6,"cacheRead":0.39,"cacheWrite":0},"contextWindow":1048572,"maxTokens":262144,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"type":"chat"},"chat:accounts/fireworks/routers/glm-flash-latest":{"id":"accounts/fireworks/routers/glm-flash-latest","name":"GLM Flash Latest (GLM 5.3 Flash)","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":0.15,"output":0.5,"cacheRead":0.03,"cacheWrite":0},"contextWindow":1048573,"maxTokens":131072,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/routers/glm-latest":{"id":"accounts/fireworks/routers/glm-latest","name":"GLM Latest","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":1.4,"output":4.4,"cacheRead":0.26,"cacheWrite":0},"contextWindow":1048573,"maxTokens":262144,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"type":"chat"},"chat:accounts/fireworks/routers/kimi-k3-fast":{"id":"accounts/fireworks/routers/kimi-k3-fast","name":"Kimi K3 Fast","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":4.5,"output":22.5,"cacheRead":0.45,"cacheWrite":0},"contextWindow":1048576,"maxTokens":131072,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false,"requiresReasoningContentOnAssistantMessages":true,"thinkingFormat":"openai","supportsMidConvoSystemMessages":true,"supportsMidConvoToolAdditions":true},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"}}}
+{"anthropic-messages":{"chat:accounts/fireworks/models/deepseek-v4p1-flash":{"id":"accounts/fireworks/models/deepseek-v4p1-flash","name":"DeepSeek V4.1 Flash","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":0.3,"output":1.2,"cacheRead":0.006,"cacheWrite":0},"contextWindow":1000000,"maxTokens":384000,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/models/ember-1":{"id":"accounts/fireworks/models/ember-1","name":"Ember-1","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":3,"output":15,"cacheRead":0.3,"cacheWrite":0},"contextWindow":1048576,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":"medium","high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/models/gpt-oss-120b":{"id":"accounts/fireworks/models/gpt-oss-120b","name":"GPT OSS 120B","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":0.15,"output":0.6,"cacheRead":0.015,"cacheWrite":0},"contextWindow":131072,"maxTokens":32768,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":"medium","high":"high","xhigh":null,"max":null},"type":"chat"},"chat:accounts/fireworks/models/inkling":{"id":"accounts/fireworks/models/inkling","name":"Inkling","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":1,"output":4.05,"cacheRead":0.17,"cacheWrite":0},"contextWindow":1048576,"maxTokens":1048576,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/models/minimax-m3":{"id":"accounts/fireworks/models/minimax-m3","name":"MiniMax-M3","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":0.3,"output":1.2,"cacheRead":0.06,"cacheWrite":0},"contextWindow":512000,"maxTokens":512000,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":"medium","high":"high","xhigh":null,"max":null},"type":"chat"},"chat:accounts/fireworks/models/muse-spark-open":{"id":"accounts/fireworks/models/muse-spark-open","name":"Muse Spark Open","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":1.25,"output":4.25,"cacheRead":0.15,"cacheWrite":0},"contextWindow":1048576,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":"medium","high":"high","xhigh":"xhigh","max":null},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/models/nemotron-3-ultra-nvfp4":{"id":"accounts/fireworks/models/nemotron-3-ultra-nvfp4","name":"Nemotron 3 Ultra 550B A55B","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":0.6,"output":2.4,"cacheRead":0.12,"cacheWrite":0},"contextWindow":262144,"maxTokens":128000,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},"type":"chat"},"chat:accounts/fireworks/models/nemotron-lightning-3p5-30b-a3b":{"id":"accounts/fireworks/models/nemotron-lightning-3p5-30b-a3b","name":"Nemotron 3.5 Lightning 30B A3B","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":0.05,"output":0.2,"cacheRead":0.01,"cacheWrite":0},"contextWindow":262144,"maxTokens":262144,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},"type":"chat"},"chat:accounts/fireworks/models/qwen3p8-2p4t-a95b":{"id":"accounts/fireworks/models/qwen3p8-2p4t-a95b","name":"Qwen3.8 2.4T A95B","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":2,"output":6,"cacheRead":0.25,"cacheWrite":0},"contextWindow":262144,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":"medium","high":null,"xhigh":"xhigh","max":null},"type":"chat"},"chat:accounts/fireworks/models/qwen3p8-max":{"id":"accounts/fireworks/models/qwen3p8-max","name":"Qwen3.8 Max","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":2,"output":6,"cacheRead":0.25,"cacheWrite":0},"contextWindow":262144,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":"medium","high":null,"xhigh":"xhigh","max":null},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/routers/deepseek-flash-latest":{"id":"accounts/fireworks/routers/deepseek-flash-latest","name":"DeepSeek Flash Latest","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":0.3,"output":1.2,"cacheRead":0.006,"cacheWrite":0},"contextWindow":1000000,"maxTokens":384000,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/routers/kimi-fast-latest":{"id":"accounts/fireworks/routers/kimi-fast-latest","name":"Kimi Fast Latest","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":4.5,"output":22.5,"cacheRead":0.45,"cacheWrite":0},"contextWindow":1048576,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":"medium","high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/routers/kimi-latest":{"id":"accounts/fireworks/routers/kimi-latest","name":"Kimi Latest","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":3,"output":15,"cacheRead":0.3,"cacheWrite":0},"contextWindow":1048576,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":"none","minimal":null,"low":"low","medium":"medium","high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/routers/minimax-latest":{"id":"accounts/fireworks/routers/minimax-latest","name":"MiniMax Latest","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":0.3,"output":1.2,"cacheRead":0.06,"cacheWrite":0},"contextWindow":512000,"maxTokens":512000,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false,"forceAdaptiveThinking":true},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":"medium","high":"high","xhigh":null,"max":null},"type":"chat"},"chat:accounts/fireworks/routers/qwen-max-latest":{"id":"accounts/fireworks/routers/qwen-max-latest","name":"Qwen Max Latest (Qwen3.8 Max)","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":2,"output":6,"cacheRead":0.25,"cacheWrite":0},"contextWindow":262144,"maxTokens":131072,"api":"anthropic-messages","baseUrl":"https://api.fireworks.ai/inference","compat":{"supportsToolReferences":true,"allowEmptySignature":true,"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"}},"openai-completions":{"chat:accounts/fireworks/models/glm-5p3":{"id":"accounts/fireworks/models/glm-5p3","name":"GLM 5.3","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":1.4,"output":4.4,"cacheRead":0.26,"cacheWrite":0},"contextWindow":1048573,"maxTokens":262144,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false},"thinkingLevelMap":{"off":"none","minimal":null,"low":null,"medium":null,"high":"high","xhigh":null,"max":"max"},"type":"chat"},"chat:accounts/fireworks/models/glm-5p3-flash":{"id":"accounts/fireworks/models/glm-5p3-flash","name":"GLM 5.3 Flash","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":0.15,"output":0.5,"cacheRead":0.03,"cacheWrite":0},"contextWindow":1048573,"maxTokens":131072,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false},"thinkingLevelMap":{"off":"none","minimal":null,"low":null,"medium":null,"high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/models/kimi-k3":{"id":"accounts/fireworks/models/kimi-k3","name":"Kimi K3","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":3,"output":15,"cacheRead":0.3,"cacheWrite":0},"contextWindow":1048576,"maxTokens":131072,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false,"requiresReasoningContentOnAssistantMessages":true,"thinkingFormat":"openai","supportsMidConvoSystemMessages":true,"supportsMidConvoToolAdditions":true},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/routers/glm-5p3-fast":{"id":"accounts/fireworks/routers/glm-5p3-fast","name":"GLM 5.3 Fast","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":2.1,"output":6.6,"cacheRead":0.39,"cacheWrite":0},"contextWindow":1048572,"maxTokens":262144,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false},"thinkingLevelMap":{"off":"none","minimal":null,"low":null,"medium":null,"high":"high","xhigh":null,"max":"max"},"type":"chat"},"chat:accounts/fireworks/routers/glm-fast-latest":{"id":"accounts/fireworks/routers/glm-fast-latest","name":"GLM 5.3 Fast (Latest)","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":2.1,"output":6.6,"cacheRead":0.39,"cacheWrite":0},"contextWindow":1048572,"maxTokens":262144,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"type":"chat"},"chat:accounts/fireworks/routers/glm-flash-latest":{"id":"accounts/fireworks/routers/glm-flash-latest","name":"GLM Flash Latest (GLM 5.3 Flash)","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":0.15,"output":0.5,"cacheRead":0.03,"cacheWrite":0},"contextWindow":1048573,"maxTokens":131072,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"},"chat:accounts/fireworks/routers/glm-latest":{"id":"accounts/fireworks/routers/glm-latest","name":"GLM Latest","provider":"fireworks","reasoning":true,"input":["text"],"cost":{"input":1.4,"output":4.4,"cacheRead":0.26,"cacheWrite":0},"contextWindow":1048573,"maxTokens":262144,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"type":"chat"},"chat:accounts/fireworks/routers/kimi-k3-fast":{"id":"accounts/fireworks/routers/kimi-k3-fast","name":"Kimi K3 Fast","provider":"fireworks","reasoning":true,"input":["text","image"],"cost":{"input":4.5,"output":22.5,"cacheRead":0.45,"cacheWrite":0},"contextWindow":1048576,"maxTokens":131072,"api":"openai-completions","baseUrl":"https://api.fireworks.ai/inference/v1","compat":{"supportsStrictMode":true,"supportsStore":false,"supportsDeveloperRole":false,"sendSessionAffinityHeaders":true,"supportsLongCacheRetention":false,"requiresReasoningContentOnAssistantMessages":true,"thinkingFormat":"openai","supportsMidConvoSystemMessages":true,"supportsMidConvoToolAdditions":true},"thinkingLevelMap":{"off":null,"minimal":null,"low":"low","medium":null,"high":"high","xhigh":null,"max":"max"},"inputLimits":{"images":{"resize":{"maxWidth":2000,"maxHeight":2000,"maxBytes":4718592,"jpegQuality":80}}},"type":"chat"}}}