smoltalk 0.15.3 → 0.17.0

package/dist/models.js

@@ -20,6 +20,7 @@ export const speechToTextModels = [
     {
         type: "speech-to-text",
         modelName: "whisper-1",
+        description: "Scheduled shutdown February 26, 2027 (announced 2026-08-26); use gpt-transcribe instead.",
         perMinuteCost: 0.006,
         provider: "openai",
         supportedMimeTypes: [
@@ -28,6 +29,13 @@ export const speechToTextModels = [
         ],
         maxBytes: 25 * 1024 * 1024,
     },
+    {
+        type: "speech-to-text",
+        modelName: "gpt-transcribe",
+        description: "OpenAI's transcription model on v1/audio/transcriptions; the named replacement for whisper-1.",
+        perMinuteCost: 0.0045, // verified 2026-09-27
+        provider: "openai",
+    },
     {
         type: "speech-to-text",
         modelName: "whisper-large-v3",
@@ -92,6 +100,24 @@ export const textToSpeechModels = [
     },
     // Gemini TTS is token-billed (text input + audio output). No maxInputChars:
     // Gemini documents a 32k-token context, and characters are not a sound proxy.
+    {
+        type: "text-to-speech",
+        modelName: "gemini-3.8-flash-tts",
+        provider: "google",
+        description: "Gemini 3.8 Flash TTS (GA 2026-09-22). Introductory pricing through December 31, 2026; $1.00 input / $18.00 audio output per 1M from January 1, 2027.",
+        inputTokenCost: 0.5, // $/1M text-input tokens, verified 2026-09-27
+        outputAudioTokenCost: 9.0, // $/1M audio-output tokens
+        formats: ["pcm", "wav"],
+    },
+    {
+        type: "text-to-speech",
+        modelName: "gemini-3.8-flash-lite-tts",
+        provider: "google",
+        description: "Gemini 3.8 Flash-Lite TTS (GA 2026-09-22). Introductory pricing through December 31, 2026; $1.00 input / $12.00 audio output per 1M from January 1, 2027.",
+        inputTokenCost: 0.5, // $/1M text-input tokens, verified 2026-09-27
+        outputAudioTokenCost: 6.0, // $/1M audio-output tokens
+        formats: ["pcm", "wav"],
+    },
     {
         type: "text-to-speech",
         modelName: "gemini-3.1-flash-tts-preview",
@@ -127,7 +153,7 @@ export const textModels = [
         inputTokenCost: 0.15,
         cachedInputTokenCost: 0.075,
         outputTokenCost: 0.6,
-        outputTokensPerSecond: 65,
+        outputTokensPerSecond: 110,
         modalities: {
             input: ["text", "image", "pdf"],
             output: ["text"],
@@ -167,7 +193,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "o3",
-        description: "o3 is a reasoning model that sets a new standard for math, science, coding, visual reasoning tasks, and technical writing. Part of the o-series of reasoning models. Knowledge cutoff: June 2024.",
+        description: "o3 is a reasoning model that sets a new standard for math, science, coding, visual reasoning tasks, and technical writing. Part of the o-series of reasoning models. Knowledge cutoff: June 2024. Scheduled shutdown December 11, 2026 (announced 2026-06-11); migrate to the gpt-5.6 family.",
         maxInputTokens: 200000,
         maxOutputTokens: 100000,
         inputTokenCost: 2,
@@ -197,7 +223,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "o3-mini",
-        description: "o3-mini is our most recent small reasoning model, providing high intelligence at the same cost and latency targets of o1-mini. o3-mini also supports key developer features, like Structured Outputs, function calling, Batch API, and more. Like other models in the o-series, it is designed to excel at science, math, and coding tasks. Knowledge cutoff: June 2024.",
+        description: "o3-mini is our most recent small reasoning model, providing high intelligence at the same cost and latency targets of o1-mini. o3-mini also supports key developer features, like Structured Outputs, function calling, Batch API, and more. Like other models in the o-series, it is designed to excel at science, math, and coding tasks. Knowledge cutoff: June 2024. Scheduled shutdown October 23, 2026; migrate to gpt-5.6-sol.",
         maxInputTokens: 200000,
         maxOutputTokens: 100000,
         inputTokenCost: 1.1,
@@ -227,7 +253,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "o4-mini",
-        description: "Latest small o-series model optimized for fast, effective reasoning with exceptional performance in coding and visual tasks. Knowledge cutoff: June 2024.",
+        description: "Latest small o-series model optimized for fast, effective reasoning with exceptional performance in coding and visual tasks. Knowledge cutoff: June 2024. Scheduled shutdown October 23, 2026; migrate to gpt-5.6-terra.",
         maxInputTokens: 200000,
         maxOutputTokens: 100000,
         inputTokenCost: 1.1,
@@ -257,7 +283,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "o3-pro",
-        description: "o3-pro uses more compute for complex reasoning tasks. Available via Responses API only. Requests may take several minutes. Knowledge cutoff: June 2024.",
+        description: "o3-pro uses more compute for complex reasoning tasks. Available via Responses API only. Requests may take several minutes. Knowledge cutoff: June 2024. Scheduled shutdown December 11, 2026 (announced 2026-06-11); migrate to gpt-5.6-sol with pro reasoning.",
         maxInputTokens: 200000,
         maxOutputTokens: 100000,
         inputTokenCost: 20,
@@ -283,7 +309,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "o1",
-        description: "o1 is a reasoning model designed to excel at complex reasoning tasks including science, math, and coding. The knowledge cutoff for o1 models is October, 2023.",
+        description: "o1 is a reasoning model designed to excel at complex reasoning tasks including science, math, and coding. The knowledge cutoff for o1 models is October, 2023. Scheduled shutdown October 23, 2026; migrate to gpt-5.6-sol.",
         maxInputTokens: 200000,
         maxOutputTokens: 100000,
         inputTokenCost: 15,
@@ -386,7 +412,7 @@ export const textModels = [
         inputTokenCost: 2,
         cachedInputTokenCost: 0.5,
         outputTokenCost: 8,
-        outputTokensPerSecond: 105,
+        outputTokensPerSecond: 141,
         modalities: {
             input: ["text", "image", "pdf"],
             output: ["text"],
@@ -409,7 +435,7 @@ export const textModels = [
         inputTokenCost: 0.4,
         cachedInputTokenCost: 0.1,
         outputTokenCost: 1.6,
-        outputTokensPerSecond: 78,
+        outputTokensPerSecond: 142,
         modalities: {
             input: ["text", "image", "pdf"],
             output: ["text"],
@@ -426,7 +452,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "gpt-4.1-nano",
-        description: "GPT-4.1 nano is the fastest and most affordable GPT-4.1 variant with 1M token context window. Knowledge cutoff: June 2024.",
+        description: "GPT-4.1 nano is the fastest and most affordable GPT-4.1 variant with 1M token context window. Knowledge cutoff: June 2024. Scheduled shutdown October 23, 2026; migrate to gpt-5.6-luna.",
         maxInputTokens: 1047576,
         maxOutputTokens: 32768,
         inputTokenCost: 0.1,
@@ -449,7 +475,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "gpt-5",
-        description: "GPT-5 is a frontier reasoning model with 400K context window. Supports reasoning tokens. Knowledge cutoff: September 2024.",
+        description: "GPT-5 is a frontier reasoning model with 400K context window. Supports reasoning tokens. Knowledge cutoff: September 2024. Scheduled shutdown December 11, 2026 (announced 2026-06-11); migrate to the gpt-5.6 family.",
         maxInputTokens: 400000,
         maxOutputTokens: 128000,
         inputTokenCost: 1.25,
@@ -479,13 +505,13 @@ export const textModels = [
     {
         type: "text",
         modelName: "gpt-5-mini",
-        description: "GPT-5 mini is a faster, more cost-efficient version of GPT-5 with 400K context window. Knowledge cutoff: May 2024.",
+        description: "GPT-5 mini is a faster, more cost-efficient version of GPT-5 with 400K context window. Knowledge cutoff: May 2024. Scheduled shutdown December 11, 2026 (announced 2026-06-11); migrate to the gpt-5.6 family.",
         maxInputTokens: 400000,
         maxOutputTokens: 128000,
         inputTokenCost: 0.25,
         cachedInputTokenCost: 0.025,
         outputTokenCost: 2,
-        outputTokensPerSecond: 69,
+        outputTokensPerSecond: 127,
         reasoning: {
             levels: ["minimal", "low", "medium", "high"],
             defaultLevel: "medium",
@@ -509,7 +535,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "gpt-5-nano",
-        description: "GPT-5 nano is the fastest and most affordable GPT-5 variant with 400K context window. Knowledge cutoff: May 2024.",
+        description: "GPT-5 nano is the fastest and most affordable GPT-5 variant with 400K context window. Knowledge cutoff: May 2024. Scheduled shutdown December 11, 2026 (announced 2026-06-11); migrate to the gpt-5.6 family.",
         maxInputTokens: 400000,
         maxOutputTokens: 128000,
         inputTokenCost: 0.05,
@@ -545,6 +571,7 @@ export const textModels = [
         inputTokenCost: 1.25,
         cachedInputTokenCost: 0.125,
         outputTokenCost: 10,
+        outputTokensPerSecond: 106,
         reasoning: {
             levels: ["none", "low", "medium", "high"],
             defaultLevel: "none",
@@ -574,7 +601,7 @@ export const textModels = [
         inputTokenCost: 1.75,
         cachedInputTokenCost: 0.175,
         outputTokenCost: 14,
-        outputTokensPerSecond: 61,
+        outputTokensPerSecond: 79,
         reasoning: {
             levels: ["none", "low", "medium", "high"],
             defaultLevel: "none",
@@ -630,6 +657,7 @@ export const textModels = [
         inputTokenCost: 2.5,
         cachedInputTokenCost: 0.25,
         outputTokenCost: 15,
+        outputTokensPerSecond: 88,
         longContext: {
             inputTokenCost: 5,
             cachedInputTokenCost: 0.5,
@@ -665,7 +693,7 @@ export const textModels = [
         inputTokenCost: 0.75,
         cachedInputTokenCost: 0.075,
         outputTokenCost: 4.5,
-        outputTokensPerSecond: 180,
+        outputTokensPerSecond: 216,
         reasoning: {
             levels: ["none", "low", "medium", "high", "xhigh"],
             defaultLevel: "none",
@@ -758,7 +786,7 @@ export const textModels = [
         inputTokenCost: 5,
         cachedInputTokenCost: 0.5,
         outputTokenCost: 30,
-        outputTokensPerSecond: 66,
+        outputTokensPerSecond: 86,
         longContext: {
             inputTokenCost: 10,
             cachedInputTokenCost: 1,
@@ -821,13 +849,13 @@ export const textModels = [
     {
         type: "text",
         modelName: "gpt-6-astra",
-        description: "GPT-6 Astra is OpenAI's most capable model, built for the hardest end-to-end work: complex reasoning, coding, computer use, research, and document creation. 1M context window. Standard pricing for ≤272K input tokens; prompts above that are billed at 2x input/cache and 1.5x output for the whole request. Knowledge cutoff: April 2026.",
+        description: "GPT-6 Astra is OpenAI's most capable model, built for the hardest end-to-end work: complex reasoning, coding, computer use, research, and document creation. 1M context window. Standard pricing for ≤272K input tokens; prompts above that are billed at 2x input/cache and 1.5x output for the whole request. Does not support custom temperature/top_p or logprobs. Knowledge cutoff: April 2026.",
         maxInputTokens: 1050000,
         maxOutputTokens: 128000,
         inputTokenCost: 10,
         cachedInputTokenCost: 1,
         outputTokenCost: 50,
-        outputTokensPerSecond: 69,
+        outputTokensPerSecond: 52,
         longContext: {
             inputTokenCost: 20,
             cachedInputTokenCost: 2,
@@ -846,8 +874,117 @@ export const textModels = [
             output: ["text"],
         },
         knowledge: "2026-04-30",
-        releaseDate: "2026-09-04",
-        lastUpdated: "2026-09-04",
+        releaseDate: "2026-09-03",
+        lastUpdated: "2026-09-03",
+        family: "gpt",
+        openWeights: false,
+        structuredOutput: true,
+        temperatureSupported: false,
+        provider: "openai",
+    },
+    {
+        type: "text",
+        modelName: "gpt-6.1-sol",
+        description: "GPT-6.1 Sol (released 2026-09-29) is the newer Sol model, replacing gpt-6-sol in OpenAI's flagship lineup at the same per-token price; cache reads are 0.05x input ($0.10/1M) instead of 0.1x. 1M context window. Standard pricing for ≤272K input tokens; prompts above that are billed at 2x input/cache and 1.5x output for the whole request. Reasoning cannot be set to none. Cataloged under openai-responses because tool calling requires the Responses API (Chat Completions works only without tools). Knowledge cutoff: April 2026.",
+        maxInputTokens: 1050000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 2,
+        cachedInputTokenCost: 0.1,
+        cacheCreationInputTokenCost: 2.5,
+        outputTokenCost: 10,
+        outputTokensPerSecond: 64,
+        longContext: {
+            inputTokenCost: 4,
+            cachedInputTokenCost: 0.2,
+            outputTokenCost: 15,
+            thresholdTokens: 272000,
+        },
+        reasoning: {
+            levels: ["low", "medium", "high", "xhigh", "max"],
+            defaultLevel: "medium",
+            canDisable: false,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        modalities: {
+            input: ["text", "image"],
+            output: ["text"],
+        },
+        knowledge: "2026-04-30",
+        releaseDate: "2026-09-29",
+        lastUpdated: "2026-09-29",
+        family: "gpt",
+        openWeights: false,
+        structuredOutput: true,
+        temperatureSupported: false,
+        provider: "openai-responses",
+    },
+    {
+        type: "text",
+        modelName: "gpt-6-sol",
+        description: "GPT-6 Sol balances capability and cost in the GPT-6 family. Superseded by gpt-6.1-sol (same price, cheaper cache reads); not deprecated. 1M context window. Standard pricing for ≤272K input tokens; prompts above that are billed at 2x input/cache and 1.5x output for the whole request. Knowledge cutoff: April 2026.",
+        maxInputTokens: 1050000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 2,
+        cachedInputTokenCost: 0.2,
+        outputTokenCost: 10,
+        outputTokensPerSecond: 87,
+        longContext: {
+            inputTokenCost: 4,
+            cachedInputTokenCost: 0.4,
+            outputTokenCost: 15.0,
+            thresholdTokens: 272000,
+        },
+        reasoning: {
+            levels: ["none", "low", "medium", "high", "xhigh", "max"],
+            defaultLevel: "medium",
+            canDisable: true,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        modalities: {
+            input: ["text", "image"],
+            output: ["text"],
+        },
+        knowledge: "2026-04-20",
+        releaseDate: "2026-09-22",
+        lastUpdated: "2026-09-22",
+        family: "gpt",
+        openWeights: false,
+        structuredOutput: true,
+        temperatureSupported: false,
+        provider: "openai",
+    },
+    {
+        type: "text",
+        modelName: "gpt-6-luna",
+        description: "GPT-6 Luna is the fast, most affordable member of the GPT-6 family. 1M context window. Standard pricing for ≤272K input tokens; prompts above that are billed at 2x input/cache and 1.5x output for the whole request. Knowledge cutoff: May 2026.",
+        maxInputTokens: 1050000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 0.1,
+        cachedInputTokenCost: 0.01,
+        outputTokenCost: 0.5,
+        outputTokensPerSecond: 128,
+        longContext: {
+            inputTokenCost: 0.2,
+            cachedInputTokenCost: 0.02,
+            outputTokenCost: 0.75,
+            thresholdTokens: 272000,
+        },
+        reasoning: {
+            levels: ["none", "low", "medium", "high", "xhigh", "max"],
+            defaultLevel: "medium",
+            canDisable: true,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        modalities: {
+            input: ["text", "image"],
+            output: ["text"],
+        },
+        knowledge: "2026-05-18",
+        releaseDate: "2026-09-22",
+        lastUpdated: "2026-09-22",
         family: "gpt",
         openWeights: false,
         structuredOutput: true,
@@ -935,7 +1072,7 @@ export const textModels = [
         inputTokenCost: 0.2,
         cachedInputTokenCost: 0.02,
         outputTokenCost: 1.2,
-        outputTokensPerSecond: 165,
+        outputTokensPerSecond: 124,
         longContext: {
             inputTokenCost: 0.4,
             cachedInputTokenCost: 0.04,
@@ -1030,14 +1167,14 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-3.8-flash",
-        description: "Google's most intelligent Flash model (GA September 2, 2026), built on Gemini 3.7 Flash and engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows. 1M context window, 64K output. Introductory pricing $0.75 input / $3.75 output per 1M through December 31, 2026 ($1.50 / $7.50 thereafter). Thinking levels low/medium/high — `minimal` returns an error.",
+        description: "Google's most intelligent Flash model (GA September 2, 2026), built on Gemini 3.7 Flash and engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows. 1M context window, 64K output. Introductory pricing $0.75 input (all modalities, including audio) / $3.75 output per 1M through December 31, 2026 ($1.50 / $7.50 thereafter). Thinking levels low/medium/high — `minimal` returns an error.",
         maxInputTokens: 1048576,
         maxOutputTokens: 65536,
         inputTokenCost: 0.75,
         cachedInputTokenCost: 0.075,
         outputTokenCost: 3.75,
-        outputTokensPerSecond: 329,
-        inputAudioTokenCost: 1.5,
+        outputTokensPerSecond: 249,
+        inputAudioTokenCost: 0.75,
         reasoning: {
             levels: ["low", "medium", "high"],
             defaultLevel: "medium",
@@ -1061,14 +1198,14 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-3.7-flash",
-        description: "Gemini 3.7 Flash (GA August 13, 2026) for complex coding, agentic workflows, and reliable multi-step execution. Superseded by gemini-3.8-flash but fully supported. 1M context window, 64K output. Introductory pricing $0.75 input / $3.75 output per 1M through December 31, 2026 ($1.50 / $7.50 thereafter). Thinking levels low/medium/high — `minimal` returns an error.",
+        description: "Gemini 3.7 Flash (GA August 13, 2026) for complex coding, agentic workflows, and reliable multi-step execution. Superseded by gemini-3.8-flash but fully supported. 1M context window, 64K output. Introductory pricing $0.75 input (all modalities, including audio) / $3.75 output per 1M through December 31, 2026 ($1.50 / $7.50 thereafter). Thinking levels low/medium/high — `minimal` returns an error.",
         maxInputTokens: 1048576,
         maxOutputTokens: 65536,
         inputTokenCost: 0.75,
         cachedInputTokenCost: 0.075,
         outputTokenCost: 3.75,
         outputTokensPerSecond: 272,
-        inputAudioTokenCost: 1.5,
+        inputAudioTokenCost: 0.75,
         reasoning: {
             levels: ["low", "medium", "high"],
             defaultLevel: "medium",
@@ -1092,16 +1229,17 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-3.6-flash",
-        description: "Gemini 3.6 Flash (GA July 21, 2026), balancing speed and multimodal capabilities across general agentic and everyday tasks. Superseded by gemini-3.7-flash / gemini-3.8-flash. 1M context window, 64K output. Introductory pricing $0.75 input / $3.75 output per 1M through December 31, 2026 ($1.50 / $7.50 thereafter). Context caching: $0.075/1M read.",
+        description: "Gemini 3.6 Flash (GA July 21, 2026), balancing speed and multimodal capabilities across general agentic and everyday tasks. Superseded by gemini-3.7-flash / gemini-3.8-flash. 1M context window, 64K output. Introductory pricing $0.75 input (all modalities, including audio) / $3.75 output per 1M through December 31, 2026 ($1.50 / $7.50 thereafter). Context caching: $0.075/1M read.",
         maxInputTokens: 1048576,
         maxOutputTokens: 65536,
         inputTokenCost: 0.75,
         cachedInputTokenCost: 0.075,
         outputTokenCost: 3.75,
-        inputAudioTokenCost: 1.5,
+        outputTokensPerSecond: 185,
+        inputAudioTokenCost: 0.75,
         reasoning: {
             levels: ["minimal", "low", "medium", "high"],
-            defaultLevel: "high",
+            defaultLevel: "medium",
             canDisable: false,
             outputsThinking: true,
             outputsSignatures: true,
@@ -1128,10 +1266,11 @@ export const textModels = [
         inputTokenCost: 1.5,
         cachedInputTokenCost: 0.15,
         outputTokenCost: 9,
+        outputTokensPerSecond: 211,
         inputAudioTokenCost: 1.5,
         reasoning: {
             levels: ["minimal", "low", "medium", "high"],
-            defaultLevel: "high",
+            defaultLevel: "medium",
             canDisable: false,
             outputsThinking: true,
             outputsSignatures: true,
@@ -1152,7 +1291,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-3-flash-preview",
-        description: "Gemini 3 Flash preview. Superseded by gemini-3.5-flash. 1M context window and 64K output. Optimized for agentic workflows and coding.",
+        description: "Gemini 3 Flash preview. Superseded by gemini-3.6-flash (Google's named replacement); no shutdown date announced. 1M context window and 64K output. Optimized for agentic workflows and coding.",
         maxInputTokens: 1048576,
         maxOutputTokens: 65536,
         inputTokenCost: 0.5,
@@ -1190,7 +1329,7 @@ export const textModels = [
         cachedInputTokenCost: 0.03,
         outputTokenCost: 2.5,
         outputTokensPerSecond: 347,
-        inputAudioTokenCost: 0.5,
+        inputAudioTokenCost: 0.3,
         reasoning: {
             levels: ["minimal", "low", "medium", "high"],
             defaultLevel: "minimal",
@@ -1214,13 +1353,13 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-3.1-flash-lite",
-        description: "Cost-effective Gemini 3.1 model (GA). Superseded by gemini-3.5-flash-lite. Thinking support, 1M context window, 64K output. 2.5x faster TTFA and 45% faster output than 2.5 Flash.",
+        description: "Cost-effective Gemini 3.1 model (GA). Superseded by gemini-3.5-flash-lite; scheduled shutdown May 7, 2027. Thinking support, 1M context window, 64K output. 2.5x faster TTFA and 45% faster output than 2.5 Flash.",
         maxInputTokens: 1048576,
         maxOutputTokens: 65536,
         inputTokenCost: 0.25,
         cachedInputTokenCost: 0.025,
         outputTokenCost: 1.5,
-        outputTokensPerSecond: 379,
+        outputTokensPerSecond: 267,
         inputAudioTokenCost: 0.5,
         reasoning: {
             levels: ["minimal", "low", "medium", "high"],
@@ -1245,7 +1384,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-3.1-flash-lite-preview",
-        description: "DEPRECATED: Preview version, discontinued July 9, 2026. Use gemini-3.1-flash-lite instead.",
+        description: "DEPRECATED: Preview version, shut down May 25, 2026. Use gemini-3.1-flash-lite instead.",
         maxInputTokens: 1048576,
         maxOutputTokens: 65536,
         inputTokenCost: 0.25,
@@ -1269,13 +1408,13 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-2.5-pro",
-        description: "High-performance Gemini 2.5 model with 1M context window. Adaptive thinking for complex reasoning and coding. Standard pricing for ≤200k tokens ($1.25 input/$10.00 output), higher rates for >200k tokens ($2.50 input/$15.00 output). Batch API: 50% discount.",
+        description: "High-performance Gemini 2.5 model with 1M context window. Adaptive thinking for complex reasoning and coding. Standard pricing for ≤200k tokens ($1.25 input/$10.00 output), higher rates for >200k tokens ($2.50 input/$15.00 output). Batch API: 50% discount. Since September 18, 2026, access is limited to projects that have used it before (not deprecated; served until further notice).",
         maxInputTokens: 1048576,
         maxOutputTokens: 65536,
         inputTokenCost: 1.25,
         cachedInputTokenCost: 0.125,
         outputTokenCost: 10,
-        outputTokensPerSecond: 145,
+        outputTokensPerSecond: 122,
         longContext: {
             inputTokenCost: 2.5,
             cachedInputTokenCost: 0.25,
@@ -1303,7 +1442,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-2.5-flash",
-        description: "Balanced Gemini 2.5 model with excellent performance-to-cost ratio. Lightning-fast with controllable thinking budgets. 1M context window. Context caching available for up to 75% cost reduction.",
+        description: "Balanced Gemini 2.5 model with excellent performance-to-cost ratio. Lightning-fast with controllable thinking budgets. 1M context window. Context caching available for up to 75% cost reduction. Since September 18, 2026, access is limited to projects that have used it before (not deprecated; served until further notice).",
         maxInputTokens: 1048576,
         maxOutputTokens: 65536,
         inputTokenCost: 0.3,
@@ -1337,13 +1476,13 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-2.5-flash-lite",
-        description: "Most cost-effective Gemini 2.5 option for high-throughput applications. 1M context window.",
+        description: "Most cost-effective Gemini 2.5 option for high-throughput applications. 1M context window. Since September 18, 2026, access is limited to projects that have used it before (not deprecated; served until further notice).",
         maxInputTokens: 1048576,
         maxOutputTokens: 65536,
         inputTokenCost: 0.1,
         cachedInputTokenCost: 0.01,
         outputTokenCost: 0.4,
-        outputTokensPerSecond: 400,
+        outputTokensPerSecond: 297,
         inputAudioTokenCost: 0.3,
         reasoning: {
             canDisable: true,
@@ -1458,10 +1597,41 @@ export const textModels = [
         disabled: true,
         provider: "google",
     },
+    {
+        type: "text",
+        modelName: "claude-opus-5-5",
+        description: "Anthropic's recommended default model (released 2026-09-22), superseding claude-opus-5 at a lower price. Cache reads are 0.05x base input ($0.20/MTok) rather than the standard 0.1x. Adaptive thinking is always on (cannot be disabled — `disabled` or `budget_tokens` returns a 400) and defaults to effort `medium`; thinking text is omitted unless requested. Forced tool use (tool_choice any/tool) returns a 400. 1M context window, 128K max output.",
+        maxInputTokens: 1000000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 4,
+        cachedInputTokenCost: 0.2,
+        cacheCreationInputTokenCost: 5,
+        outputTokenCost: 20,
+        outputTokensPerSecond: 92,
+        reasoning: {
+            thinkingStyle: "adaptive",
+            levels: ["low", "medium", "high", "xhigh", "max"],
+            defaultLevel: "medium",
+            canDisable: false,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
+        modalities: {
+            input: ["text", "image", "pdf"],
+            output: ["text"],
+        },
+        knowledge: "2026-06",
+        releaseDate: "2026-09-22",
+        lastUpdated: "2026-09-22",
+        family: "claude-opus",
+        openWeights: false,
+        temperatureSupported: false,
+        provider: "anthropic",
+    },
     {
         type: "text",
         modelName: "claude-fable-5-1",
-        description: "Anthropic's most capable widely released model, for the most demanding reasoning and long-horizon agentic work. Successor to claude-fable-5 in the same tier at the same per-token price, but cache reads are 0.025x base input ($0.25/MTok) rather than the standard 0.1x. Thinking is always on (cannot be disabled); the raw chain of thought is never returned. Forced tool use (tool_choice any/tool) returns a 400. Requires 30-day data retention (not available under zero data retention). 1M context window, 128K max output.",
+        description: "Anthropic's most capable widely released model, for the most demanding reasoning and long-horizon agentic work (claude-opus-5-5 is the recommended default for everything else). Successor to claude-fable-5 in the same tier at the same per-token price, but cache reads are 0.025x base input ($0.25/MTok) rather than the standard 0.1x. Thinking is always on (cannot be disabled); the raw chain of thought is never returned. Forced tool use (tool_choice any/tool) returns a 400. Requires 30-day data retention (not available under zero data retention). 1M context window, 128K max output.",
         maxInputTokens: 1000000,
         maxOutputTokens: 128000,
         inputTokenCost: 10,
@@ -1492,7 +1662,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "claude-opus-5",
-        description: "Opus-tier model above Opus 4.8, strongest on deep reasoning, agentic and long-horizon work, at half the cost of claude-fable-5-1. Thinking is on by default; it can be disabled only at effort `high` or below (xhigh/max return a 400). 1M context window, 128K max output.",
+        description: "Legacy Opus-tier model, superseded by claude-opus-5-5 (which is cheaper at $4/$20). Strong on deep reasoning, agentic and long-horizon work. Thinking is on by default; it can be disabled only at effort `high` or below (xhigh/max return a 400). 1M context window, 128K max output.",
         maxInputTokens: 1000000,
         maxOutputTokens: 128000,
         inputTokenCost: 5,
@@ -1554,16 +1724,19 @@ export const textModels = [
     {
         type: "text",
         modelName: "claude-opus-4-8",
-        description: "The most capable Claude model for complex reasoning and agentic coding. Same per-token pricing as Opus 4.7 with improved tool-use efficiency (~290 tokens for tool-use system prompt vs 675 on 4.7). 1M context window, 128K max output.",
+        description: "Claude Opus 4.8 — legacy model; prefer claude-opus-5-5. Same per-token pricing as Opus 4.7 with improved tool-use efficiency (~290 tokens for tool-use system prompt vs 675 on 4.7). Adaptive thinking is off by default. 1M context window, 128K max output.",
         maxInputTokens: 1000000,
         maxOutputTokens: 128000,
         inputTokenCost: 5,
         cachedInputTokenCost: 0.5,
         cacheCreationInputTokenCost: 6.25,
         outputTokenCost: 25,
+        outputTokensPerSecond: 61,
         reasoning: {
             thinkingStyle: "adaptive",
-            canDisable: false,
+            levels: ["low", "medium", "high", "xhigh", "max"],
+            defaultLevel: "high",
+            canDisable: true,
             outputsThinking: true,
             outputsSignatures: true,
         },
@@ -1571,6 +1744,7 @@ export const textModels = [
             input: ["text", "image", "pdf"],
             output: ["text"],
         },
+        knowledge: "2026-01",
         releaseDate: "2026-05-28",
         lastUpdated: "2026-05-28",
         family: "claude-opus",
@@ -1588,10 +1762,12 @@ export const textModels = [
         cachedInputTokenCost: 0.5,
         cacheCreationInputTokenCost: 6.25,
         outputTokenCost: 25,
-        outputTokensPerSecond: 72,
+        outputTokensPerSecond: 48,
         reasoning: {
             thinkingStyle: "adaptive",
-            canDisable: false,
+            levels: ["low", "medium", "high", "xhigh", "max"],
+            defaultLevel: "high",
+            canDisable: true,
             outputsThinking: true,
             outputsSignatures: true,
         },
@@ -1610,16 +1786,18 @@ export const textModels = [
     {
         type: "text",
         modelName: "claude-opus-4-6",
-        description: "Claude Opus 4.6 — legacy model. Use claude-opus-4-7 instead. 1M context window, 128K max output.",
+        description: "Claude Opus 4.6 — legacy model. Use claude-opus-5-5 instead. 1M context window, 128K max output.",
         maxInputTokens: 1000000,
         maxOutputTokens: 128000,
         inputTokenCost: 5,
         cachedInputTokenCost: 0.5,
         cacheCreationInputTokenCost: 6.25,
         outputTokenCost: 25,
-        outputTokensPerSecond: 53,
+        outputTokensPerSecond: 38,
         reasoning: {
             thinkingStyle: "adaptive",
+            levels: ["low", "medium", "high", "max"],
+            defaultLevel: "high",
             canDisable: true,
             outputsThinking: true,
             outputsSignatures: true,
@@ -1636,10 +1814,40 @@ export const textModels = [
         temperatureSupported: true,
         provider: "anthropic",
     },
+    {
+        type: "text",
+        modelName: "claude-sonnet-5-5",
+        description: "Current Sonnet-tier model (released 2026-09-28), superseding claude-sonnet-5 at the same price. Adaptive thinking is always on and defaults to effort `high`; `thinking: {type: \"disabled\"}` and `budget_tokens` return a 400. The lowest setting is `thinking: {type: \"between_tools\"}` (low/medium/high effort only), which skips up-front thinking but still emits short thinking blocks between tool calls. Thinking text is omitted unless requested. Forced tool use (tool_choice any/tool) and non-default temperature/top_p/top_k return a 400. Same tokenizer as Sonnet 5. 1M context window, 128K max output.",
+        maxInputTokens: 1000000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 2,
+        cachedInputTokenCost: 0.2,
+        cacheCreationInputTokenCost: 2.5,
+        outputTokenCost: 10,
+        reasoning: {
+            thinkingStyle: "adaptive",
+            levels: ["low", "medium", "high", "xhigh", "max"],
+            defaultLevel: "high",
+            canDisable: false,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
+        modalities: {
+            input: ["text", "image", "pdf"],
+            output: ["text"],
+        },
+        knowledge: "2026-06",
+        releaseDate: "2026-09-28",
+        lastUpdated: "2026-09-28",
+        family: "claude-sonnet",
+        openWeights: false,
+        temperatureSupported: false,
+        provider: "anthropic",
+    },
     {
         type: "text",
         modelName: "claude-sonnet-5",
-        description: "The best combination of speed and intelligence in the Sonnet tier, with near-Opus quality on coding and agentic work. Adaptive thinking on by default; supports the full low/medium/high/xhigh/max effort range. New tokenizer (~30% more tokens for the same text vs Sonnet 4.6). The $2/$10 per MTok launch rate is now the standard price — the increase to $3/$15 scheduled for 2026-09-01 was cancelled. 1M context window, 128K max output.",
+        description: "Legacy Sonnet-tier model, superseded by claude-sonnet-5-5 at the same price. Near-Opus quality on coding and agentic work. Adaptive thinking on by default; supports the full low/medium/high/xhigh/max effort range. New tokenizer (~30% more tokens for the same text vs Sonnet 4.6). The $2/$10 per MTok launch rate is now the standard price — the increase to $3/$15 scheduled for 2026-09-01 was cancelled. 1M context window, 128K max output.",
         maxInputTokens: 1000000,
         maxOutputTokens: 128000,
         inputTokenCost: 2,
@@ -1680,6 +1888,8 @@ export const textModels = [
         outputTokensPerSecond: 52,
         reasoning: {
             thinkingStyle: "adaptive",
+            levels: ["low", "medium", "high", "max"],
+            defaultLevel: "high",
             canDisable: true,
             outputsThinking: true,
             outputsSignatures: true,
@@ -1765,6 +1975,7 @@ export const textModels = [
         cachedInputTokenCost: 0.1,
         cacheCreationInputTokenCost: 1.25,
         outputTokenCost: 5,
+        outputTokensPerSecond: 97,
         reasoning: {
             thinkingStyle: "budget",
             canDisable: true,
@@ -1786,13 +1997,14 @@ export const textModels = [
     {
         type: "text",
         modelName: "claude-opus-4-5",
-        description: "Claude Opus 4.5 — earlier Opus generation. Prefer claude-opus-4-8.",
+        description: "Claude Opus 4.5 — earlier Opus generation. Prefer claude-opus-5-5. Supports the effort parameter alongside budget_tokens thinking.",
         maxInputTokens: 200000,
         maxOutputTokens: 64000,
         inputTokenCost: 5,
         cachedInputTokenCost: 0.5,
         cacheCreationInputTokenCost: 6.25,
         outputTokenCost: 25,
+        outputTokensPerSecond: 49,
         reasoning: {
             thinkingStyle: "budget",
             canDisable: true,
@@ -1814,13 +2026,14 @@ export const textModels = [
     {
         type: "text",
         modelName: "claude-sonnet-4-5",
-        description: "Claude Sonnet 4.5 — earlier Sonnet generation. Prefer claude-sonnet-4-6.",
+        description: "Claude Sonnet 4.5 — earlier Sonnet generation. Deprecated 2026-09-30; retires November 30, 2026. Migrate to claude-sonnet-5-5.",
         maxInputTokens: 200000,
         maxOutputTokens: 64000,
         inputTokenCost: 3,
         cachedInputTokenCost: 0.3,
         cacheCreationInputTokenCost: 3.75,
         outputTokenCost: 15,
+        outputTokensPerSecond: 46,
         reasoning: {
             thinkingStyle: "budget",
             canDisable: true,
@@ -1842,7 +2055,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "gpt-5-pro",
-        description: "GPT-5 Pro uses more compute for complex reasoning tasks. Available via the Responses API.",
+        description: "GPT-5 Pro uses more compute for complex reasoning tasks. Available via the Responses API. Scheduled shutdown December 11, 2026 (announced 2026-06-11); migrate to gpt-5.6-sol with pro reasoning.",
         maxInputTokens: 400000,
         maxOutputTokens: 272000,
         inputTokenCost: 15,
@@ -1884,6 +2097,7 @@ export const imageModels = [
         type: "image",
         modelName: "gpt-image-1",
         provider: "openai",
+        description: "Scheduled shutdown October 23, 2026; migrate to gpt-image-2.5-sunburst or gpt-image-2.5-flare.",
         // Token-based pricing per https://platform.openai.com/docs/models/gpt-image-1
         inputTokenCost: 5,
         cachedInputTokenCost: 1.25,
@@ -1892,12 +2106,43 @@ export const imageModels = [
         // Rough per-image fallback estimate when usage is unavailable.
         costPerImage: 0.25,
     },
+    {
+        type: "image",
+        modelName: "gpt-image-2",
+        provider: "openai",
+        description: "Replacement for gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest (those shut down 2026-12-01).",
+        inputTokenCost: 5,
+        cachedInputTokenCost: 1.25,
+        inputImageTokenCost: 8,
+        outputImageTokenCost: 30,
+    },
+    {
+        type: "image",
+        modelName: "gpt-image-2.5-sunburst",
+        provider: "openai",
+        description: "GPT Image 2.5 Sunburst (released 2026-09-08). Quality low/medium/high/xhigh/max/auto. Cached image input $2/1M; text output is not billed.",
+        inputTokenCost: 5,
+        cachedInputTokenCost: 1.25,
+        inputImageTokenCost: 8,
+        outputImageTokenCost: 30,
+    },
+    {
+        type: "image",
+        modelName: "gpt-image-2.5-flare",
+        provider: "openai",
+        description: "GPT Image 2.5 Flare (released 2026-09-08). Quality low/medium/high/xhigh/max/auto. Cached image input $2/1M; text output is not billed.",
+        inputTokenCost: 5,
+        cachedInputTokenCost: 1.25,
+        inputImageTokenCost: 8,
+        outputImageTokenCost: 30,
+    },
     {
         type: "image",
         modelName: "gemini-2.5-flash-image",
         provider: "google",
-        description: "aka nano-banana. Graduated from preview (was gemini-2.5-flash-image-preview).",
-        costPerImage: 0.04,
+        description: "aka nano-banana. RETIRED: shut down October 2, 2026. Use gemini-3.1-flash-image or gemini-3.1-flash-lite-image.",
+        costPerImage: 0.039,
+        disabled: true,
     },
     {
         type: "image",
@@ -1910,8 +2155,16 @@ export const imageModels = [
         type: "image",
         modelName: "gemini-3-pro-image-preview",
         provider: "google",
-        description: "High-fidelity image generation with reasoning-enhanced composition. Supports legible text rendering, complex multi-turn editing, and character consistency using up to 14 reference inputs.",
+        description: "DEPRECATED: Preview version, shut down June 25, 2026. Use gemini-3-pro-image instead.",
         costPerImage: 0.05,
+        disabled: true,
+    },
+    {
+        type: "image",
+        modelName: "gemini-3-pro-image",
+        provider: "google",
+        description: "aka Nano Banana Pro (GA 2026-05-28). High-fidelity image generation with reasoning-enhanced composition. Supports legible text rendering, complex multi-turn editing, and character consistency using up to 14 reference inputs. $0.134/image at 1K or 2K, $0.24 at 4K; text input $2/1M, text/thinking output $12/1M.",
+        costPerImage: 0.134,
     },
     {
         type: "image",
@@ -1932,7 +2185,7 @@ export const imageModels = [
         type: "image",
         modelName: "gemini-3.1-flash-lite-image",
         provider: "google",
-        description: "aka Nano Banana 2 Lite (GA 2026-06-30). Fastest, most cost-effective Gemini image model (~4s generation). ~$0.034/image at 1K. Recommended replacement for gemini-2.5-flash-image.",
+        description: "aka Nano Banana 2 Lite (GA 2026-06-30). Fastest, most cost-effective Gemini image model (~4s generation). ~$0.034/image at 1K.",
         costPerImage: 0.034,
     },
 ];
@@ -1960,6 +2213,13 @@ export const embeddingsModels = [
         modelName: "gemini-embedding-2-preview",
         provider: "google",
         tokenCost: 0.2,
+        disabled: true, // shut down 2026-08-10; use gemini-embedding-2
+    },
+    {
+        type: "embeddings",
+        modelName: "gemini-embedding-2",
+        provider: "google",
+        tokenCost: 0.2,
     },
 ];
 export const decisionModels = [
@@ -2019,7 +2279,7 @@ export const hostedTools = [
         category: "web_search",
         description: "Server-side web search (Responses API).",
         providerToolId: "web_search",
-        pricing: { unit: "per_call", amount: 0.01, note: "$10/1k standard; $25/1k preview non-reasoning; plus ~8k input tokens per call. Varies by search_context_size." },
+        pricing: { unit: "per_call", amount: 0.01, note: "$10/1k standard; $25/1k preview non-reasoning (search tokens free); otherwise search tokens billed at model rates." },
     },
     {
         name: "file_search",
@@ -2035,15 +2295,15 @@ export const hostedTools = [
         category: "code_execution",
         description: "Run Python in a sandboxed container.",
         providerToolId: "code_interpreter",
-        pricing: { unit: "per_session", amount: 0.03, note: "Per container session." },
+        pricing: { unit: "per_session", amount: 0.03, note: "$0.03 per 20-minute session (1 GB container); $0.12 for 4 GB, $0.48 for 16 GB, $1.92 for 64 GB." },
     },
     {
         name: "image_generation",
         provider: "openai-responses",
         category: "image_generation",
-        description: "Generate images with gpt-image-1 as a tool.",
+        description: "Generate images with a GPT Image model as a tool.",
         providerToolId: "image_generation",
-        pricing: { unit: "tokens", note: "gpt-image-1 token pricing: $5/$10/$40 per 1M text-in/image-in/image-out." },
+        pricing: { unit: "tokens", note: "Billed at the image model's token rates: gpt-image-2.5-sunburst/flare $5/$8/$30 per 1M text-in/image-in/image-out; gpt-image-1 (shuts down 2026-10-23) $5/$10/$40." },
     },
     {
         name: "google_search",
@@ -2054,7 +2314,7 @@ export const hostedTools = [
         pricing: {
             unit: "per_call",
             amount: 0.014,
-            freeAllowance: "5,000 grounded prompts/month (Gemini 3)",
+            freeAllowance: "5,000 search requests/month (shared across Gemini 3 models)",
             note: "Billed per query; a request may issue multiple queries.",
             perModel: {
                 "gemini-2.5-pro": { amount: 0.035, freeAllowance: "1,500/day shared", note: "Billed per prompt (Gemini 2.5)." },
@@ -2083,26 +2343,31 @@ export const hostedTools = [
         name: "maps_grounding",
         provider: "google",
         category: "maps_grounding",
-        description: "Grounding with Google Maps (Gemini 3 only).",
+        description: "Grounding with Google Maps (Gemini 3 and Gemini 2.5).",
         providerToolId: "google_maps",
-        models: ["gemini-3-pro-preview", "gemini-3.1-pro-preview", "gemini-3-flash-preview", "gemini-3.5-flash", "gemini-3.6-flash", "gemini-3.7-flash", "gemini-3.8-flash", "gemini-3.1-flash-lite", "gemini-3.5-flash-lite"],
+        models: ["gemini-3-pro-preview", "gemini-3.1-pro-preview", "gemini-3-flash-preview", "gemini-3.5-flash", "gemini-3.6-flash", "gemini-3.7-flash", "gemini-3.8-flash", "gemini-3.1-flash-lite", "gemini-3.5-flash-lite", "gemini-2.5-pro", "gemini-2.5-flash", "gemini-2.5-flash-lite"],
         pricing: {
             unit: "per_call",
             amount: 0.014,
             freeAllowance: "5,000 grounded prompts/month (Gemini 3)",
             note: "$14 per 1,000 search queries on the Gemini 3 family.",
+            perModel: {
+                "gemini-2.5-pro": { amount: 0.025, freeAllowance: "10,000/day", note: "$25 per 1,000 grounded prompts (Gemini 2.5)." },
+                "gemini-2.5-flash": { amount: 0.025, freeAllowance: "1,500/day", note: "$25 per 1,000 grounded prompts (Gemini 2.5)." },
+                "gemini-2.5-flash-lite": { amount: 0.025, freeAllowance: "1,500/day", note: "$25 per 1,000 grounded prompts (Gemini 2.5)." },
+            },
         },
     },
     {
         name: "web_search",
         provider: "openrouter",
         category: "web_search",
-        description: "Web search via OpenRouter's web plugin (Exa-backed). Adds citations as annotations.",
+        description: "Web search via OpenRouter's web plugin (native search where the model supports it, otherwise Exa). Adds citations as annotations. OpenRouter has deprecated the plugin in favor of the openrouter:web_search server tool.",
         providerToolId: "web",
         pricing: {
             unit: "per_call",
-            amount: 0.02,
-            note: "$4 per 1,000 results returned by the web plugin × default 5 results/call = $0.02/call. Plus token cost for inserted context. If you override max_results, scale this accordingly.",
+            amount: 0.007,
+            note: "Exa auto: $0.007 per request, including up to 10 results; $0.001 per additional result. Exa Deep $0.012, Deep Reasoning $0.015. Plus token cost for inserted context. Native search is billed at the model provider's own rate.",
         },
     },
 ];