smoltalk 0.11.1 → 0.12.1

package/dist/model.js

@@ -1,4 +1,4 @@
-import { getModel, getModelForProvider, isSpeechToTextModel, isTextToSpeechModel, ModelNameSchema, } from "./models.js";
+import { getModel, resolveModelForProvider, isSpeechToTextModel, isTextToSpeechModel, ModelNameSchema, } from "./models.js";
 import { SmolError } from "./smolError.js";
 import { round } from "./util/util.js";
 const TOKEN_COST_UNIT = 1_000_000;
@@ -27,7 +27,7 @@ export class Model {
     calculateCost(usage) {
         let model;
         if (this.provider !== undefined) {
-            model = getModelForProvider(this.provider, this.model, this.modelData);
+            model = resolveModelForProvider(this.provider, this.model, this.modelData);
         }
         else {
             model = getModel(this.model, this.modelData);

package/dist/models.d.ts

@@ -868,16 +868,17 @@ export declare const textModels: readonly [{
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5.6-sol";
-    readonly description: "GPT-5.6 Sol is the flagship model of the GPT-5.6 family for the most complex coding and agentic tasks. 1M context window. Standard pricing for ≤272K tokens, 2x input/1.5x output for >272K. Knowledge cutoff: February 2026.";
+    readonly description: "GPT-5.6 Sol is the flagship model of the GPT-5.6 family for the most complex coding and agentic tasks. 1M context window. Standard pricing for ≤272K tokens, 2x input/1.5x output for >272K. Reflects the August 21, 2026 promotional price cut (input $5→$4, output $30→$20), which runs through at least November 21, 2026. Knowledge cutoff: February 2026.";
     readonly maxInputTokens: 1050000;
     readonly maxOutputTokens: 128000;
-    readonly inputTokenCost: 5;
-    readonly cachedInputTokenCost: 0.5;
-    readonly outputTokenCost: 30;
+    readonly inputTokenCost: 4;
+    readonly cachedInputTokenCost: 0.4;
+    readonly outputTokenCost: 20;
+    readonly outputTokensPerSecond: 71;
     readonly longContext: {
-        readonly inputTokenCost: 10;
-        readonly cachedInputTokenCost: 1;
-        readonly outputTokenCost: 45;
+        readonly inputTokenCost: 8;
+        readonly cachedInputTokenCost: 0.8;
+        readonly outputTokenCost: 30;
         readonly thresholdTokens: 200000;
     };
     readonly reasoning: {
@@ -908,6 +909,7 @@ export declare const textModels: readonly [{
     readonly inputTokenCost: 2;
     readonly cachedInputTokenCost: 0.2;
     readonly outputTokenCost: 12;
+    readonly outputTokensPerSecond: 98;
     readonly longContext: {
         readonly inputTokenCost: 4;
         readonly cachedInputTokenCost: 0.4;
@@ -942,6 +944,7 @@ export declare const textModels: readonly [{
     readonly inputTokenCost: 0.2;
     readonly cachedInputTokenCost: 0.02;
     readonly outputTokenCost: 1.2;
+    readonly outputTokensPerSecond: 116;
     readonly longContext: {
         readonly inputTokenCost: 0.4;
         readonly cachedInputTokenCost: 0.04;
@@ -1030,15 +1033,75 @@ export declare const textModels: readonly [{
     readonly temperatureSupported: true;
     readonly disabled: true;
     readonly provider: "google";
+}, {
+    readonly type: "text";
+    readonly modelName: "gemini-3.8-flash";
+    readonly 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.";
+    readonly maxInputTokens: 1048576;
+    readonly maxOutputTokens: 65536;
+    readonly inputTokenCost: 0.75;
+    readonly cachedInputTokenCost: 0.075;
+    readonly outputTokenCost: 3.75;
+    readonly outputTokensPerSecond: 312;
+    readonly inputAudioTokenCost: 1.5;
+    readonly reasoning: {
+        readonly levels: readonly ["low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
+    readonly modalities: {
+        readonly input: readonly ["text", "image", "video", "audio", "pdf"];
+        readonly output: readonly ["text"];
+    };
+    readonly knowledge: "2026-03";
+    readonly releaseDate: "2026-09-02";
+    readonly lastUpdated: "2026-09-02";
+    readonly family: "gemini-flash";
+    readonly openWeights: false;
+    readonly structuredOutput: true;
+    readonly temperatureSupported: true;
+    readonly provider: "google";
+}, {
+    readonly type: "text";
+    readonly modelName: "gemini-3.7-flash";
+    readonly 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.";
+    readonly maxInputTokens: 1048576;
+    readonly maxOutputTokens: 65536;
+    readonly inputTokenCost: 0.75;
+    readonly cachedInputTokenCost: 0.075;
+    readonly outputTokenCost: 3.75;
+    readonly outputTokensPerSecond: 272;
+    readonly inputAudioTokenCost: 1.5;
+    readonly reasoning: {
+        readonly levels: readonly ["low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
+    readonly modalities: {
+        readonly input: readonly ["text", "image", "video", "audio", "pdf"];
+        readonly output: readonly ["text"];
+    };
+    readonly knowledge: "2026-03";
+    readonly releaseDate: "2026-08-13";
+    readonly lastUpdated: "2026-08-13";
+    readonly family: "gemini-flash";
+    readonly openWeights: false;
+    readonly structuredOutput: true;
+    readonly temperatureSupported: true;
+    readonly provider: "google";
 }, {
     readonly type: "text";
     readonly modelName: "gemini-3.6-flash";
-    readonly description: "Latest Gemini 3.6 Flash model (GA July 21, 2026). Supersedes Gemini 3.5 Flash with better token efficiency and agentic planning at a lower output price ($7.50 vs $9.00/1M). 1M context window, 64K output. Context caching: $0.15/1M read.";
+    readonly 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.";
     readonly maxInputTokens: 1048576;
     readonly maxOutputTokens: 65536;
-    readonly inputTokenCost: 1.5;
-    readonly cachedInputTokenCost: 0.15;
-    readonly outputTokenCost: 7.5;
+    readonly inputTokenCost: 0.75;
+    readonly cachedInputTokenCost: 0.075;
+    readonly outputTokenCost: 3.75;
     readonly inputAudioTokenCost: 1.5;
     readonly reasoning: {
         readonly levels: readonly ["minimal", "low", "medium", "high"];
@@ -1127,6 +1190,7 @@ export declare const textModels: readonly [{
     readonly inputTokenCost: 0.3;
     readonly cachedInputTokenCost: 0.03;
     readonly outputTokenCost: 2.5;
+    readonly outputTokensPerSecond: 347;
     readonly inputAudioTokenCost: 0.5;
     readonly reasoning: {
         readonly levels: readonly ["minimal", "low", "medium", "high"];
@@ -1380,16 +1444,75 @@ export declare const textModels: readonly [{
     readonly costUnit: "characters";
     readonly disabled: true;
     readonly provider: "google";
+}, {
+    readonly type: "text";
+    readonly modelName: "claude-fable-5-1";
+    readonly 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.";
+    readonly maxInputTokens: 1000000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 10;
+    readonly cachedInputTokenCost: 0.25;
+    readonly cacheCreationInputTokenCost: 12.5;
+    readonly outputTokenCost: 50;
+    readonly reasoning: {
+        readonly thinkingStyle: "adaptive";
+        readonly levels: readonly ["low", "medium", "high", "xhigh", "max"];
+        readonly defaultLevel: "high";
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
+    readonly modalities: {
+        readonly input: readonly ["text", "image", "pdf"];
+        readonly output: readonly ["text"];
+    };
+    readonly knowledge: "2026-06";
+    readonly releaseDate: "2026-09-01";
+    readonly lastUpdated: "2026-09-01";
+    readonly family: "claude-fable";
+    readonly openWeights: false;
+    readonly temperatureSupported: false;
+    readonly provider: "anthropic";
+}, {
+    readonly type: "text";
+    readonly modelName: "claude-opus-5";
+    readonly 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.";
+    readonly maxInputTokens: 1000000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 5;
+    readonly cachedInputTokenCost: 0.5;
+    readonly cacheCreationInputTokenCost: 6.25;
+    readonly outputTokenCost: 25;
+    readonly reasoning: {
+        readonly thinkingStyle: "adaptive";
+        readonly levels: readonly ["low", "medium", "high", "xhigh", "max"];
+        readonly defaultLevel: "high";
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
+    readonly modalities: {
+        readonly input: readonly ["text", "image", "pdf"];
+        readonly output: readonly ["text"];
+    };
+    readonly knowledge: "2026-05";
+    readonly releaseDate: "2026-07-24";
+    readonly lastUpdated: "2026-07-24";
+    readonly family: "claude-opus";
+    readonly openWeights: false;
+    readonly temperatureSupported: false;
+    readonly provider: "anthropic";
 }, {
     readonly type: "text";
     readonly modelName: "claude-fable-5";
-    readonly description: "Anthropic's most capable widely released model, for the most demanding reasoning and long-horizon agentic work. Thinking is always on (cannot be disabled); the raw chain of thought is never returned. Requires 30-day data retention (not available under zero data retention). 1M context window, 128K max output.";
+    readonly description: "Previous-generation Fable model, superseded by claude-fable-5-1 at the same per-token price (5.1 also cuts cache reads to $0.25/MTok). Thinking is always on (cannot be disabled); the raw chain of thought is never returned. Requires 30-day data retention (not available under zero data retention). 1M context window, 128K max output.";
     readonly maxInputTokens: 1000000;
     readonly maxOutputTokens: 128000;
     readonly inputTokenCost: 10;
     readonly cachedInputTokenCost: 1;
     readonly cacheCreationInputTokenCost: 12.5;
     readonly outputTokenCost: 50;
+    readonly outputTokensPerSecond: 65;
     readonly reasoning: {
         readonly thinkingStyle: "adaptive";
         readonly levels: readonly ["low", "medium", "high", "xhigh", "max"];
@@ -1494,13 +1617,14 @@ export declare const textModels: readonly [{
 }, {
     readonly type: "text";
     readonly modelName: "claude-sonnet-5";
-    readonly 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). Standard pricing $3/$15; introductory $2/$10 per MTok through 2026-08-31. 1M context window, 128K max output.";
+    readonly 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.";
     readonly maxInputTokens: 1000000;
     readonly maxOutputTokens: 128000;
-    readonly inputTokenCost: 3;
-    readonly cachedInputTokenCost: 0.3;
-    readonly cacheCreationInputTokenCost: 3.75;
-    readonly outputTokenCost: 15;
+    readonly inputTokenCost: 2;
+    readonly cachedInputTokenCost: 0.2;
+    readonly cacheCreationInputTokenCost: 2.5;
+    readonly outputTokenCost: 10;
+    readonly outputTokensPerSecond: 71;
     readonly reasoning: {
         readonly thinkingStyle: "adaptive";
         readonly levels: readonly ["low", "medium", "high", "xhigh", "max"];
@@ -1805,6 +1929,13 @@ export declare function getModel(modelName: ModelName, requestData?: ModelDataBl
  * matching entry comes first and can silently pick the wrong provider.
  */
 export declare function getModelForProvider(provider: string, modelName: ModelName, requestData?: ModelDataBlob): ModelType | undefined;
+/**
+ * Like `getModelForProvider`, but when the exact provider key misses and the
+ * provider is a known API variant (e.g. "openai-responses"), retries under
+ * its catalog family ("openai"). The exact match always wins so
+ * variant-exclusive entries (o3-pro etc.) keep their own data.
+ */
+export declare function resolveModelForProvider(provider: string, modelName: ModelName, requestData?: ModelDataBlob): ModelType | undefined;
 /**
  * Whether a model is known to accept the given input modality ("image", "pdf", …).
  * Returns undefined when the model is unknown or carries no `modalities` data —

package/dist/models.js

@@ -812,16 +812,17 @@ export const textModels = [
     {
         type: "text",
         modelName: "gpt-5.6-sol",
-        description: "GPT-5.6 Sol is the flagship model of the GPT-5.6 family for the most complex coding and agentic tasks. 1M context window. Standard pricing for ≤272K tokens, 2x input/1.5x output for >272K. Knowledge cutoff: February 2026.",
+        description: "GPT-5.6 Sol is the flagship model of the GPT-5.6 family for the most complex coding and agentic tasks. 1M context window. Standard pricing for ≤272K tokens, 2x input/1.5x output for >272K. Reflects the August 21, 2026 promotional price cut (input $5→$4, output $30→$20), which runs through at least November 21, 2026. Knowledge cutoff: February 2026.",
         maxInputTokens: 1050000,
         maxOutputTokens: 128000,
-        inputTokenCost: 5,
-        cachedInputTokenCost: 0.5,
-        outputTokenCost: 30,
+        inputTokenCost: 4,
+        cachedInputTokenCost: 0.4,
+        outputTokenCost: 20,
+        outputTokensPerSecond: 71,
         longContext: {
-            inputTokenCost: 10,
-            cachedInputTokenCost: 1,
-            outputTokenCost: 45,
+            inputTokenCost: 8,
+            cachedInputTokenCost: 0.8,
+            outputTokenCost: 30,
             thresholdTokens: 200000,
         },
         reasoning: {
@@ -853,6 +854,7 @@ export const textModels = [
         inputTokenCost: 2,
         cachedInputTokenCost: 0.2,
         outputTokenCost: 12,
+        outputTokensPerSecond: 98,
         longContext: {
             inputTokenCost: 4,
             cachedInputTokenCost: 0.4,
@@ -888,6 +890,7 @@ export const textModels = [
         inputTokenCost: 0.2,
         cachedInputTokenCost: 0.02,
         outputTokenCost: 1.2,
+        outputTokensPerSecond: 116,
         longContext: {
             inputTokenCost: 0.4,
             cachedInputTokenCost: 0.04,
@@ -979,15 +982,77 @@ export const textModels = [
         disabled: true,
         provider: "google",
     },
+    {
+        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.",
+        maxInputTokens: 1048576,
+        maxOutputTokens: 65536,
+        inputTokenCost: 0.75,
+        cachedInputTokenCost: 0.075,
+        outputTokenCost: 3.75,
+        outputTokensPerSecond: 312,
+        inputAudioTokenCost: 1.5,
+        reasoning: {
+            levels: ["low", "medium", "high"],
+            defaultLevel: "medium",
+            canDisable: false,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
+        modalities: {
+            input: ["text", "image", "video", "audio", "pdf"],
+            output: ["text"],
+        },
+        knowledge: "2026-03",
+        releaseDate: "2026-09-02",
+        lastUpdated: "2026-09-02",
+        family: "gemini-flash",
+        openWeights: false,
+        structuredOutput: true,
+        temperatureSupported: true,
+        provider: "google",
+    },
+    {
+        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.",
+        maxInputTokens: 1048576,
+        maxOutputTokens: 65536,
+        inputTokenCost: 0.75,
+        cachedInputTokenCost: 0.075,
+        outputTokenCost: 3.75,
+        outputTokensPerSecond: 272,
+        inputAudioTokenCost: 1.5,
+        reasoning: {
+            levels: ["low", "medium", "high"],
+            defaultLevel: "medium",
+            canDisable: false,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
+        modalities: {
+            input: ["text", "image", "video", "audio", "pdf"],
+            output: ["text"],
+        },
+        knowledge: "2026-03",
+        releaseDate: "2026-08-13",
+        lastUpdated: "2026-08-13",
+        family: "gemini-flash",
+        openWeights: false,
+        structuredOutput: true,
+        temperatureSupported: true,
+        provider: "google",
+    },
     {
         type: "text",
         modelName: "gemini-3.6-flash",
-        description: "Latest Gemini 3.6 Flash model (GA July 21, 2026). Supersedes Gemini 3.5 Flash with better token efficiency and agentic planning at a lower output price ($7.50 vs $9.00/1M). 1M context window, 64K output. Context caching: $0.15/1M 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 / $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: 1.5,
-        cachedInputTokenCost: 0.15,
-        outputTokenCost: 7.5,
+        inputTokenCost: 0.75,
+        cachedInputTokenCost: 0.075,
+        outputTokenCost: 3.75,
         inputAudioTokenCost: 1.5,
         reasoning: {
             levels: ["minimal", "low", "medium", "high"],
@@ -1079,6 +1144,7 @@ export const textModels = [
         inputTokenCost: 0.3,
         cachedInputTokenCost: 0.03,
         outputTokenCost: 2.5,
+        outputTokensPerSecond: 347,
         inputAudioTokenCost: 0.5,
         reasoning: {
             levels: ["minimal", "low", "medium", "high"],
@@ -1347,16 +1413,77 @@ export const textModels = [
         disabled: true,
         provider: "google",
     },
+    {
+        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.",
+        maxInputTokens: 1000000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 10,
+        cachedInputTokenCost: 0.25,
+        cacheCreationInputTokenCost: 12.5,
+        outputTokenCost: 50,
+        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: "2026-09-01",
+        lastUpdated: "2026-09-01",
+        family: "claude-fable",
+        openWeights: false,
+        temperatureSupported: false,
+        provider: "anthropic",
+    },
+    {
+        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.",
+        maxInputTokens: 1000000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 5,
+        cachedInputTokenCost: 0.5,
+        cacheCreationInputTokenCost: 6.25,
+        outputTokenCost: 25,
+        reasoning: {
+            thinkingStyle: "adaptive",
+            levels: ["low", "medium", "high", "xhigh", "max"],
+            defaultLevel: "high",
+            canDisable: true,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
+        modalities: {
+            input: ["text", "image", "pdf"],
+            output: ["text"],
+        },
+        knowledge: "2026-05",
+        releaseDate: "2026-07-24",
+        lastUpdated: "2026-07-24",
+        family: "claude-opus",
+        openWeights: false,
+        temperatureSupported: false,
+        provider: "anthropic",
+    },
     {
         type: "text",
         modelName: "claude-fable-5",
-        description: "Anthropic's most capable widely released model, for the most demanding reasoning and long-horizon agentic work. Thinking is always on (cannot be disabled); the raw chain of thought is never returned. Requires 30-day data retention (not available under zero data retention). 1M context window, 128K max output.",
+        description: "Previous-generation Fable model, superseded by claude-fable-5-1 at the same per-token price (5.1 also cuts cache reads to $0.25/MTok). Thinking is always on (cannot be disabled); the raw chain of thought is never returned. Requires 30-day data retention (not available under zero data retention). 1M context window, 128K max output.",
         maxInputTokens: 1000000,
         maxOutputTokens: 128000,
         inputTokenCost: 10,
         cachedInputTokenCost: 1,
         cacheCreationInputTokenCost: 12.5,
         outputTokenCost: 50,
+        outputTokensPerSecond: 65,
         reasoning: {
             thinkingStyle: "adaptive",
             levels: ["low", "medium", "high", "xhigh", "max"],
@@ -1465,13 +1592,14 @@ export const textModels = [
     {
         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). Standard pricing $3/$15; introductory $2/$10 per MTok through 2026-08-31. 1M context window, 128K max output.",
+        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.",
         maxInputTokens: 1000000,
         maxOutputTokens: 128000,
-        inputTokenCost: 3,
-        cachedInputTokenCost: 0.3,
-        cacheCreationInputTokenCost: 3.75,
-        outputTokenCost: 15,
+        inputTokenCost: 2,
+        cachedInputTokenCost: 0.2,
+        cacheCreationInputTokenCost: 2.5,
+        outputTokenCost: 10,
+        outputTokensPerSecond: 71,
         reasoning: {
             thinkingStyle: "adaptive",
             levels: ["low", "medium", "high", "xhigh", "max"],
@@ -1884,7 +2012,7 @@ export const hostedTools = [
         category: "maps_grounding",
         description: "Grounding with Google Maps (Gemini 3 only).",
         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-3.1-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"],
         pricing: { unit: "per_call", note: "Gemini 3 family only; see Google pricing." },
     },
     {
@@ -1954,6 +2082,35 @@ export function getModel(modelName, requestData) {
 export function getModelForProvider(provider, modelName, requestData) {
     return getAllModels(requestData).find((model) => model.modelName === modelName && model.provider === provider);
 }
+/**
+ * API-variant provider names mapped to the catalog family whose entries they
+ * share. "openai-responses" is the same vendor with the same pricing and
+ * modalities as "openai" — only Responses-exclusive models are cataloged
+ * under "openai-responses" itself. Unrelated providers (gateways like
+ * "openrouter"/"openai-compat", custom registrations) are deliberately NOT
+ * aliased: a same-named model under another provider must not lend its
+ * pricing or capabilities across providers.
+ */
+const PROVIDER_FAMILY_ALIAS = {
+    "openai-responses": "openai",
+};
+/**
+ * Like `getModelForProvider`, but when the exact provider key misses and the
+ * provider is a known API variant (e.g. "openai-responses"), retries under
+ * its catalog family ("openai"). The exact match always wins so
+ * variant-exclusive entries (o3-pro etc.) keep their own data.
+ */
+export function resolveModelForProvider(provider, modelName, requestData) {
+    const exact = getModelForProvider(provider, modelName, requestData);
+    if (exact) {
+        return exact;
+    }
+    const family = PROVIDER_FAMILY_ALIAS[provider];
+    if (family === undefined) {
+        return undefined;
+    }
+    return getModelForProvider(family, modelName, requestData);
+}
 /**
  * Whether a model is known to accept the given input modality ("image", "pdf", …).
  * Returns undefined when the model is unknown or carries no `modalities` data —
@@ -1962,7 +2119,7 @@ export function getModelForProvider(provider, modelName, requestData) {
 export function modelSupportsInputModality(modelName, modality, requestData, provider) {
     let model;
     if (provider !== undefined) {
-        model = getModelForProvider(provider, modelName, requestData);
+        model = resolveModelForProvider(provider, modelName, requestData);
     }
     else {
         model = getModel(modelName, requestData);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "smoltalk",
-  "version": "0.11.1",
+  "version": "0.12.1",
   "description": "A common interface for LLM APIs",
   "homepage": "https://github.com/egonSchiele/smoltalk",
   "files": [