smoltalk 0.2.2 → 0.3.1

package/dist/models.d.ts

@@ -27,6 +27,8 @@ export type SpeechToTextModel = BaseModel & {
 export type ImageModel = BaseModel & {
     type: "image";
     costPerImage?: number;
+    inputImageTokenCost?: number;
+    outputImageTokenCost?: number;
     outputType?: "FileOutput" | "Array";
 };
 export type TextModel = BaseModel & {
@@ -50,6 +52,7 @@ export type TextModel = BaseModel & {
 export type EmbeddingsModel = {
     type: "embeddings";
     modelName: string;
+    provider: string;
     tokenCost?: number;
 };
 export type ModelType = SpeechToTextModel | TextModel | EmbeddingsModel | ImageModel;
@@ -350,6 +353,42 @@ export declare const textModels: readonly [{
         readonly outputsSignatures: false;
     };
     readonly provider: "openai";
+}, {
+    readonly type: "text";
+    readonly modelName: "gpt-5.4-mini";
+    readonly description: "GPT-5.4 mini is the strongest mini model for coding, computer use, and subagents. 400K context window. Knowledge cutoff: August 2025.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 0.75;
+    readonly cachedInputTokenCost: 0.075;
+    readonly outputTokenCost: 4.5;
+    readonly outputTokensPerSecond: 180;
+    readonly reasoning: {
+        readonly levels: readonly ["none", "low", "medium", "high", "xhigh"];
+        readonly defaultLevel: "none";
+        readonly canDisable: true;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+}, {
+    readonly type: "text";
+    readonly modelName: "gpt-5.4-nano";
+    readonly description: "GPT-5.4 nano is the fastest and most affordable GPT-5.4 variant for classification, data extraction, ranking, and sub-agents. 400K context window. Knowledge cutoff: August 2025.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 0.2;
+    readonly cachedInputTokenCost: 0.02;
+    readonly outputTokenCost: 1.25;
+    readonly outputTokensPerSecond: 160;
+    readonly reasoning: {
+        readonly levels: readonly ["none", "low", "medium", "high", "xhigh"];
+        readonly defaultLevel: "none";
+        readonly canDisable: true;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5.4-pro";
@@ -366,6 +405,40 @@ export declare const textModels: readonly [{
         readonly outputsSignatures: false;
     };
     readonly provider: "openai";
+}, {
+    readonly type: "text";
+    readonly modelName: "gpt-5.5";
+    readonly description: "GPT-5.5 is a new class of intelligence for coding and professional work. 1M context window. Standard pricing for ≤272K tokens, 2x input/1.5x output for >272K. Knowledge cutoff: December 2025.";
+    readonly maxInputTokens: 1050000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 5;
+    readonly cachedInputTokenCost: 0.5;
+    readonly outputTokenCost: 30;
+    readonly outputTokensPerSecond: 66;
+    readonly reasoning: {
+        readonly levels: readonly ["none", "low", "medium", "high", "xhigh"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: true;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+}, {
+    readonly type: "text";
+    readonly modelName: "gpt-5.5-pro";
+    readonly description: "GPT-5.5 Pro uses more compute for complex reasoning tasks. 1M context window. Knowledge cutoff: December 2025.";
+    readonly maxInputTokens: 1050000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 30;
+    readonly outputTokenCost: 180;
+    readonly reasoning: {
+        readonly levels: readonly ["none", "low", "medium", "high", "xhigh"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: true;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
 }, {
     readonly type: "text";
     readonly modelName: "gemini-3.1-pro-preview";
@@ -374,7 +447,7 @@ export declare const textModels: readonly [{
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 2;
     readonly outputTokenCost: 12;
-    readonly outputTokensPerSecond: 112;
+    readonly outputTokensPerSecond: 133;
     readonly reasoning: {
         readonly levels: readonly ["low", "medium", "high"];
         readonly defaultLevel: "high";
@@ -401,7 +474,7 @@ export declare const textModels: readonly [{
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 0.5;
     readonly outputTokenCost: 3;
-    readonly outputTokensPerSecond: 146;
+    readonly outputTokensPerSecond: 184;
     readonly reasoning: {
         readonly levels: readonly ["minimal", "low", "medium", "high"];
         readonly defaultLevel: "high";
@@ -435,7 +508,7 @@ export declare const textModels: readonly [{
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 1.25;
     readonly outputTokenCost: 10;
-    readonly outputTokensPerSecond: 134;
+    readonly outputTokensPerSecond: 145;
     readonly reasoning: {
         readonly canDisable: false;
         readonly outputsThinking: true;
@@ -538,12 +611,28 @@ export declare const textModels: readonly [{
     readonly costUnit: "characters";
     readonly disabled: true;
     readonly provider: "google";
+}, {
+    readonly type: "text";
+    readonly modelName: "claude-opus-4-7";
+    readonly description: "The most capable Claude model for complex reasoning and agentic coding. Features Adaptive Thinking that auto-tunes reasoning compute per request. 1M context window, 128K max output. Knowledge cutoff: January 2026.";
+    readonly maxInputTokens: 1000000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 5;
+    readonly cachedInputTokenCost: 0.5;
+    readonly outputTokenCost: 25;
+    readonly outputTokensPerSecond: 72;
+    readonly reasoning: {
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
+    readonly provider: "anthropic";
 }, {
     readonly type: "text";
     readonly modelName: "claude-opus-4-6";
-    readonly description: "The most intelligent Claude model for building agents and coding. 200K context window (1M in beta), 128K max output.";
-    readonly maxInputTokens: 200000;
-    readonly maxOutputTokens: 131072;
+    readonly description: "Claude Opus 4.6 — legacy model. Use claude-opus-4-7 instead. 1M context window, 128K max output.";
+    readonly maxInputTokens: 1000000;
+    readonly maxOutputTokens: 128000;
     readonly inputTokenCost: 5;
     readonly cachedInputTokenCost: 0.5;
     readonly outputTokenCost: 25;
@@ -557,12 +646,13 @@ export declare const textModels: readonly [{
 }, {
     readonly type: "text";
     readonly modelName: "claude-sonnet-4-6";
-    readonly description: "The best combination of speed and intelligence. 200K context window (1M in beta), 64K max output.";
-    readonly maxInputTokens: 200000;
+    readonly description: "The best combination of speed and intelligence. 1M context window, 64K max output.";
+    readonly maxInputTokens: 1000000;
     readonly maxOutputTokens: 64000;
     readonly inputTokenCost: 3;
     readonly cachedInputTokenCost: 0.3;
     readonly outputTokenCost: 15;
+    readonly outputTokensPerSecond: 52;
     readonly reasoning: {
         readonly canDisable: true;
         readonly outputsThinking: true;
@@ -617,12 +707,22 @@ export declare const imageModels: readonly [{
     readonly type: "image";
     readonly modelName: "gpt-image-1";
     readonly provider: "openai";
+    readonly inputTokenCost: 5;
+    readonly cachedInputTokenCost: 1.25;
+    readonly inputImageTokenCost: 10;
+    readonly outputImageTokenCost: 40;
     readonly costPerImage: 0.25;
 }, {
     readonly type: "image";
-    readonly modelName: "gemini-2.5-flash-image-preview";
+    readonly modelName: "gemini-2.5-flash-image";
+    readonly provider: "google";
+    readonly description: "aka nano-banana. Graduated from preview (was gemini-2.5-flash-image-preview).";
+    readonly costPerImage: 0.04;
+}, {
+    readonly type: "image";
+    readonly modelName: "nano-banana-pro-preview";
     readonly provider: "google";
-    readonly description: "aka nano-banana";
+    readonly description: "Higher-tier nano-banana. Preview model.";
     readonly costPerImage: 0.04;
 }, {
     readonly type: "image";
@@ -630,12 +730,14 @@ export declare const imageModels: readonly [{
     readonly provider: "google";
     readonly 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.";
     readonly costPerImage: 0.05;
+}, {
+    readonly type: "image";
+    readonly modelName: "gemini-3.1-flash-image-preview";
+    readonly provider: "google";
+    readonly description: "Fast image generation with Gemini 3.1 Flash. Supports resolutions from 512px to 4096px. ~$0.067/image at 1K resolution.";
+    readonly costPerImage: 0.067;
 }];
-export declare const embeddingsModels: {
-    type: string;
-    modelName: string;
-    tokenCost: number;
-}[];
+export declare const embeddingsModels: EmbeddingsModel[];
 export type TextModelName = (typeof textModels)[number]["modelName"];
 export type ImageModelName = (typeof imageModels)[number]["modelName"];
 export type SpeechToTextModelName = (typeof speechToTextModels)[number]["modelName"];
@@ -645,7 +747,7 @@ export declare const registeredTextModels: TextModel[];
 export declare function registerTextModel(model: Omit<TextModel, "type"> & {
     type?: "text";
 }): void;
-export declare function getModel(modelName: ModelName): TextModel | {
+export declare function getModel(modelName: ModelName): TextModel | EmbeddingsModel | {
     readonly type: "speech-to-text";
     readonly modelName: "whisper-web";
     readonly perMinuteCost: 0.006;
@@ -941,6 +1043,42 @@ export declare function getModel(modelName: ModelName): TextModel | {
         readonly outputsSignatures: false;
     };
     readonly provider: "openai";
+} | {
+    readonly type: "text";
+    readonly modelName: "gpt-5.4-mini";
+    readonly description: "GPT-5.4 mini is the strongest mini model for coding, computer use, and subagents. 400K context window. Knowledge cutoff: August 2025.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 0.75;
+    readonly cachedInputTokenCost: 0.075;
+    readonly outputTokenCost: 4.5;
+    readonly outputTokensPerSecond: 180;
+    readonly reasoning: {
+        readonly levels: readonly ["none", "low", "medium", "high", "xhigh"];
+        readonly defaultLevel: "none";
+        readonly canDisable: true;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+} | {
+    readonly type: "text";
+    readonly modelName: "gpt-5.4-nano";
+    readonly description: "GPT-5.4 nano is the fastest and most affordable GPT-5.4 variant for classification, data extraction, ranking, and sub-agents. 400K context window. Knowledge cutoff: August 2025.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 0.2;
+    readonly cachedInputTokenCost: 0.02;
+    readonly outputTokenCost: 1.25;
+    readonly outputTokensPerSecond: 160;
+    readonly reasoning: {
+        readonly levels: readonly ["none", "low", "medium", "high", "xhigh"];
+        readonly defaultLevel: "none";
+        readonly canDisable: true;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
 } | {
     readonly type: "text";
     readonly modelName: "gpt-5.4-pro";
@@ -957,6 +1095,40 @@ export declare function getModel(modelName: ModelName): TextModel | {
         readonly outputsSignatures: false;
     };
     readonly provider: "openai";
+} | {
+    readonly type: "text";
+    readonly modelName: "gpt-5.5";
+    readonly description: "GPT-5.5 is a new class of intelligence for coding and professional work. 1M context window. Standard pricing for ≤272K tokens, 2x input/1.5x output for >272K. Knowledge cutoff: December 2025.";
+    readonly maxInputTokens: 1050000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 5;
+    readonly cachedInputTokenCost: 0.5;
+    readonly outputTokenCost: 30;
+    readonly outputTokensPerSecond: 66;
+    readonly reasoning: {
+        readonly levels: readonly ["none", "low", "medium", "high", "xhigh"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: true;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+} | {
+    readonly type: "text";
+    readonly modelName: "gpt-5.5-pro";
+    readonly description: "GPT-5.5 Pro uses more compute for complex reasoning tasks. 1M context window. Knowledge cutoff: December 2025.";
+    readonly maxInputTokens: 1050000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 30;
+    readonly outputTokenCost: 180;
+    readonly reasoning: {
+        readonly levels: readonly ["none", "low", "medium", "high", "xhigh"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: true;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
 } | {
     readonly type: "text";
     readonly modelName: "gemini-3.1-pro-preview";
@@ -965,7 +1137,7 @@ export declare function getModel(modelName: ModelName): TextModel | {
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 2;
     readonly outputTokenCost: 12;
-    readonly outputTokensPerSecond: 112;
+    readonly outputTokensPerSecond: 133;
     readonly reasoning: {
         readonly levels: readonly ["low", "medium", "high"];
         readonly defaultLevel: "high";
@@ -992,7 +1164,7 @@ export declare function getModel(modelName: ModelName): TextModel | {
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 0.5;
     readonly outputTokenCost: 3;
-    readonly outputTokensPerSecond: 146;
+    readonly outputTokensPerSecond: 184;
     readonly reasoning: {
         readonly levels: readonly ["minimal", "low", "medium", "high"];
         readonly defaultLevel: "high";
@@ -1026,7 +1198,7 @@ export declare function getModel(modelName: ModelName): TextModel | {
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 1.25;
     readonly outputTokenCost: 10;
-    readonly outputTokensPerSecond: 134;
+    readonly outputTokensPerSecond: 145;
     readonly reasoning: {
         readonly canDisable: false;
         readonly outputsThinking: true;
@@ -1129,12 +1301,28 @@ export declare function getModel(modelName: ModelName): TextModel | {
     readonly costUnit: "characters";
     readonly disabled: true;
     readonly provider: "google";
+} | {
+    readonly type: "text";
+    readonly modelName: "claude-opus-4-7";
+    readonly description: "The most capable Claude model for complex reasoning and agentic coding. Features Adaptive Thinking that auto-tunes reasoning compute per request. 1M context window, 128K max output. Knowledge cutoff: January 2026.";
+    readonly maxInputTokens: 1000000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 5;
+    readonly cachedInputTokenCost: 0.5;
+    readonly outputTokenCost: 25;
+    readonly outputTokensPerSecond: 72;
+    readonly reasoning: {
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
+    readonly provider: "anthropic";
 } | {
     readonly type: "text";
     readonly modelName: "claude-opus-4-6";
-    readonly description: "The most intelligent Claude model for building agents and coding. 200K context window (1M in beta), 128K max output.";
-    readonly maxInputTokens: 200000;
-    readonly maxOutputTokens: 131072;
+    readonly description: "Claude Opus 4.6 — legacy model. Use claude-opus-4-7 instead. 1M context window, 128K max output.";
+    readonly maxInputTokens: 1000000;
+    readonly maxOutputTokens: 128000;
     readonly inputTokenCost: 5;
     readonly cachedInputTokenCost: 0.5;
     readonly outputTokenCost: 25;
@@ -1148,12 +1336,13 @@ export declare function getModel(modelName: ModelName): TextModel | {
 } | {
     readonly type: "text";
     readonly modelName: "claude-sonnet-4-6";
-    readonly description: "The best combination of speed and intelligence. 200K context window (1M in beta), 64K max output.";
-    readonly maxInputTokens: 200000;
+    readonly description: "The best combination of speed and intelligence. 1M context window, 64K max output.";
+    readonly maxInputTokens: 1000000;
     readonly maxOutputTokens: 64000;
     readonly inputTokenCost: 3;
     readonly cachedInputTokenCost: 0.3;
     readonly outputTokenCost: 15;
+    readonly outputTokensPerSecond: 52;
     readonly reasoning: {
         readonly canDisable: true;
         readonly outputsThinking: true;
@@ -1207,12 +1396,22 @@ export declare function getModel(modelName: ModelName): TextModel | {
     readonly type: "image";
     readonly modelName: "gpt-image-1";
     readonly provider: "openai";
+    readonly inputTokenCost: 5;
+    readonly cachedInputTokenCost: 1.25;
+    readonly inputImageTokenCost: 10;
+    readonly outputImageTokenCost: 40;
     readonly costPerImage: 0.25;
 } | {
     readonly type: "image";
-    readonly modelName: "gemini-2.5-flash-image-preview";
+    readonly modelName: "gemini-2.5-flash-image";
+    readonly provider: "google";
+    readonly description: "aka nano-banana. Graduated from preview (was gemini-2.5-flash-image-preview).";
+    readonly costPerImage: 0.04;
+} | {
+    readonly type: "image";
+    readonly modelName: "nano-banana-pro-preview";
     readonly provider: "google";
-    readonly description: "aka nano-banana";
+    readonly description: "Higher-tier nano-banana. Preview model.";
     readonly costPerImage: 0.04;
 } | {
     readonly type: "image";
@@ -1220,6 +1419,12 @@ export declare function getModel(modelName: ModelName): TextModel | {
     readonly provider: "google";
     readonly 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.";
     readonly costPerImage: 0.05;
+} | {
+    readonly type: "image";
+    readonly modelName: "gemini-3.1-flash-image-preview";
+    readonly provider: "google";
+    readonly description: "Fast image generation with Gemini 3.1 Flash. Supports resolutions from 512px to 4096px. ~$0.067/image at 1K resolution.";
+    readonly costPerImage: 0.067;
 } | undefined;
 export declare function isImageModel(model: ModelType): model is ImageModel;
 export declare function isTextModel(model: ModelType): model is TextModel;

package/dist/models.js

@@ -339,6 +339,44 @@ export const textModels = [
         },
         provider: "openai",
     },
+    {
+        type: "text",
+        modelName: "gpt-5.4-mini",
+        description: "GPT-5.4 mini is the strongest mini model for coding, computer use, and subagents. 400K context window. Knowledge cutoff: August 2025.",
+        maxInputTokens: 400000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 0.75,
+        cachedInputTokenCost: 0.075,
+        outputTokenCost: 4.5,
+        outputTokensPerSecond: 180,
+        reasoning: {
+            levels: ["none", "low", "medium", "high", "xhigh"],
+            defaultLevel: "none",
+            canDisable: true,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        provider: "openai",
+    },
+    {
+        type: "text",
+        modelName: "gpt-5.4-nano",
+        description: "GPT-5.4 nano is the fastest and most affordable GPT-5.4 variant for classification, data extraction, ranking, and sub-agents. 400K context window. Knowledge cutoff: August 2025.",
+        maxInputTokens: 400000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 0.2,
+        cachedInputTokenCost: 0.02,
+        outputTokenCost: 1.25,
+        outputTokensPerSecond: 160,
+        reasoning: {
+            levels: ["none", "low", "medium", "high", "xhigh"],
+            defaultLevel: "none",
+            canDisable: true,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        provider: "openai",
+    },
     {
         type: "text",
         modelName: "gpt-5.4-pro",
@@ -356,6 +394,42 @@ export const textModels = [
         },
         provider: "openai",
     },
+    {
+        type: "text",
+        modelName: "gpt-5.5",
+        description: "GPT-5.5 is a new class of intelligence for coding and professional work. 1M context window. Standard pricing for ≤272K tokens, 2x input/1.5x output for >272K. Knowledge cutoff: December 2025.",
+        maxInputTokens: 1_050_000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 5,
+        cachedInputTokenCost: 0.5,
+        outputTokenCost: 30,
+        outputTokensPerSecond: 66,
+        reasoning: {
+            levels: ["none", "low", "medium", "high", "xhigh"],
+            defaultLevel: "medium",
+            canDisable: true,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        provider: "openai",
+    },
+    {
+        type: "text",
+        modelName: "gpt-5.5-pro",
+        description: "GPT-5.5 Pro uses more compute for complex reasoning tasks. 1M context window. Knowledge cutoff: December 2025.",
+        maxInputTokens: 1_050_000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 30,
+        outputTokenCost: 180,
+        reasoning: {
+            levels: ["none", "low", "medium", "high", "xhigh"],
+            defaultLevel: "medium",
+            canDisable: true,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        provider: "openai",
+    },
     {
         type: "text",
         modelName: "gemini-3.1-pro-preview",
@@ -364,7 +438,7 @@ export const textModels = [
         maxOutputTokens: 65536,
         inputTokenCost: 2.0,
         outputTokenCost: 12.0,
-        outputTokensPerSecond: 112,
+        outputTokensPerSecond: 133,
         reasoning: {
             levels: ["low", "medium", "high"],
             defaultLevel: "high",
@@ -393,7 +467,7 @@ export const textModels = [
         maxOutputTokens: 65536,
         inputTokenCost: 0.5,
         outputTokenCost: 3.0,
-        outputTokensPerSecond: 146,
+        outputTokensPerSecond: 184,
         reasoning: {
             levels: ["minimal", "low", "medium", "high"],
             defaultLevel: "high",
@@ -429,7 +503,7 @@ export const textModels = [
         maxOutputTokens: 65536,
         inputTokenCost: 1.25,
         outputTokenCost: 10.0,
-        outputTokensPerSecond: 134,
+        outputTokensPerSecond: 145,
         reasoning: {
             canDisable: false,
             outputsThinking: true,
@@ -541,12 +615,29 @@ export const textModels = [
         disabled: true,
         provider: "google",
     },
+    {
+        type: "text",
+        modelName: "claude-opus-4-7",
+        description: "The most capable Claude model for complex reasoning and agentic coding. Features Adaptive Thinking that auto-tunes reasoning compute per request. 1M context window, 128K max output. Knowledge cutoff: January 2026.",
+        maxInputTokens: 1_000_000,
+        maxOutputTokens: 128_000,
+        inputTokenCost: 5,
+        cachedInputTokenCost: 0.5,
+        outputTokenCost: 25,
+        outputTokensPerSecond: 72,
+        reasoning: {
+            canDisable: false,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
+        provider: "anthropic",
+    },
     {
         type: "text",
         modelName: "claude-opus-4-6",
-        description: "The most intelligent Claude model for building agents and coding. 200K context window (1M in beta), 128K max output.",
-        maxInputTokens: 200_000,
-        maxOutputTokens: 131_072,
+        description: "Claude Opus 4.6 — legacy model. Use claude-opus-4-7 instead. 1M context window, 128K max output.",
+        maxInputTokens: 1_000_000,
+        maxOutputTokens: 128_000,
         inputTokenCost: 5,
         cachedInputTokenCost: 0.5,
         outputTokenCost: 25,
@@ -561,12 +652,13 @@ export const textModels = [
     {
         type: "text",
         modelName: "claude-sonnet-4-6",
-        description: "The best combination of speed and intelligence. 200K context window (1M in beta), 64K max output.",
-        maxInputTokens: 200_000,
+        description: "The best combination of speed and intelligence. 1M context window, 64K max output.",
+        maxInputTokens: 1_000_000,
         maxOutputTokens: 64_000,
         inputTokenCost: 3,
         cachedInputTokenCost: 0.3,
         outputTokenCost: 15,
+        outputTokensPerSecond: 52,
         reasoning: {
             canDisable: true,
             outputsThinking: true,
@@ -626,14 +718,26 @@ export const imageModels = [
         type: "image",
         modelName: "gpt-image-1",
         provider: "openai",
-        // varies: https://platform.openai.com/docs/models/gpt-image-1
+        // Token-based pricing per https://platform.openai.com/docs/models/gpt-image-1
+        inputTokenCost: 5,
+        cachedInputTokenCost: 1.25,
+        inputImageTokenCost: 10,
+        outputImageTokenCost: 40,
+        // Rough per-image fallback estimate when usage is unavailable.
         costPerImage: 0.25,
     },
     {
         type: "image",
-        modelName: "gemini-2.5-flash-image-preview",
+        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,
+    },
+    {
+        type: "image",
+        modelName: "nano-banana-pro-preview",
         provider: "google",
-        description: "aka nano-banana",
+        description: "Higher-tier nano-banana. Preview model.",
         costPerImage: 0.04,
     },
     {
@@ -643,9 +747,39 @@ export const imageModels = [
         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.",
         costPerImage: 0.05,
     },
+    {
+        type: "image",
+        modelName: "gemini-3.1-flash-image-preview",
+        provider: "google",
+        description: "Fast image generation with Gemini 3.1 Flash. Supports resolutions from 512px to 4096px. ~$0.067/image at 1K resolution.",
+        costPerImage: 0.067,
+    },
 ];
 export const embeddingsModels = [
-    { type: "embeddings", modelName: "text-embedding-3-small", tokenCost: 0.02 },
+    {
+        type: "embeddings",
+        modelName: "text-embedding-3-small",
+        provider: "openai",
+        tokenCost: 0.02,
+    },
+    {
+        type: "embeddings",
+        modelName: "text-embedding-3-large",
+        provider: "openai",
+        tokenCost: 0.13,
+    },
+    {
+        type: "embeddings",
+        modelName: "gemini-embedding-001",
+        provider: "google",
+        tokenCost: 0.15,
+    },
+    {
+        type: "embeddings",
+        modelName: "gemini-embedding-2-preview",
+        provider: "google",
+        tokenCost: 0.2,
+    },
 ];
 export const registeredTextModels = [];
 export function registerTextModel(model) {
@@ -657,6 +791,7 @@ export function getModel(modelName) {
         ...imageModels,
         ...speechToTextModels,
         ...registeredTextModels,
+        ...embeddingsModels,
     ];
     return allModels.find((model) => model.modelName === modelName);
 }