smoltalk 0.0.43 → 0.0.45

package/dist/clients/google.d.ts

@@ -1,8 +1,13 @@
-import { GoogleGenAI } from "@google/genai";
+import { Content, GenerateContentConfig, GoogleGenAI } from "@google/genai";
 import { BaseClientConfig, PromptConfig, PromptResult, Result, SmolClient, StreamChunk } from "../types.js";
 import { BaseClient } from "./baseClient.js";
 import { ModelName } from "../models.js";
 export type SmolGoogleConfig = BaseClientConfig;
+type GeneratedRequest = {
+    contents: Content[];
+    model: ModelName;
+    config: GenerateContentConfig;
+};
 export declare class SmolGoogle extends BaseClient implements SmolClient {
     private client;
     private logger;
@@ -13,5 +18,7 @@ export declare class SmolGoogle extends BaseClient implements SmolClient {
     private calculateUsageAndCost;
     private buildRequest;
     _textSync(config: PromptConfig): Promise<Result<PromptResult>>;
+    __textSync(request: GeneratedRequest): Promise<Result<PromptResult>>;
     _textStream(config: PromptConfig): AsyncGenerator<StreamChunk>;
 }
+export {};

package/dist/clients/google.js

@@ -1,10 +1,11 @@
 import { GoogleGenAI } from "@google/genai";
 import { ToolCall } from "../classes/ToolCall.js";
 import { getLogger } from "../logger.js";
-import { success, } from "../types.js";
+import { addCosts, addTokenUsage, success, } from "../types.js";
 import { zodToGoogleTool } from "../util/tool.js";
 import { BaseClient } from "./baseClient.js";
 import { Model } from "../model.js";
+import { userMessage } from "../classes/message/index.js";
 export class SmolGoogle extends BaseClient {
     client;
     logger;
@@ -65,10 +66,7 @@ export class SmolGoogle extends BaseClient {
         if (tools.length > 0) {
             genConfig.tools = [{ functionDeclarations: tools }];
         }
-        if (config.responseFormat && tools.length > 0) {
-            console.error("Warning: Both responseFormat and tools are specified in the prompt config. Google Gemini does not support enforcing a response format when tools are included, so the responseFormat will be ignored.");
-        }
-        else if (config.responseFormat && tools.length === 0) {
+        if (config.responseFormat) {
             genConfig.responseMimeType = "application/json";
             genConfig.responseJsonSchema = config.responseFormat.toJSONSchema();
         }
@@ -94,6 +92,67 @@ export class SmolGoogle extends BaseClient {
         if (signal) {
             request.config = { ...request.config, abortSignal: signal };
         }
+        const hasTools = config.tools && config.tools.length > 0;
+        const hasStructuredResponse = !!config.responseFormat;
+        if (!hasTools && !hasStructuredResponse) {
+            // If there are no tools or structured response, we can make a single request and return immediately
+            return this.__textSync(request);
+        }
+        // Google Gemini does not support combining function calling with
+        // responseMimeType 'application/json'. When tools are present, we
+        // make two requests instead
+        /*********** TOOL CALL REQUEST ************/
+        this.logger.debug("Detected both tool calls and structured response in call to Google Gemini. Making separate request to Google Gemini for tool calls.");
+        const toolRequest = {
+            ...request,
+            config: {
+                ...request.config,
+                responseMimeType: undefined,
+                responseJsonSchema: undefined,
+            },
+        };
+        const toolResult = await this.__textSync(toolRequest);
+        if (!toolResult.success) {
+            return toolResult;
+        }
+        if (toolResult.value.toolCalls.length > 0) {
+            this.logger.debug("Tool calls detected. Returning tool calls without making second request for structured response.");
+            return toolResult;
+        }
+        if (!toolResult.value.output) {
+            throw new Error("No output or tool calls detected in Google Gemini response. This should not happen.");
+        }
+        this.logger.debug("No tool calls detected. Making second request to Google Gemini for structured response.");
+        /*********** STRUCTURED OUTPUT REQUEST ************/
+        const message = userMessage(`Please return this output in the specified structured format. Output: ${toolResult.value.output}`);
+        const messages = [message.toGoogleMessage()];
+        const responseRequest = {
+            ...request,
+            config: {
+                ...request.config,
+                tools: undefined,
+            },
+            messages,
+        };
+        const responseResult = await this.__textSync(responseRequest);
+        if (!responseResult.success) {
+            return responseResult;
+        }
+        const thinkingBlocks = [
+            ...(toolResult.value.thinkingBlocks || []),
+            ...(responseResult.value.thinkingBlocks || []),
+        ];
+        return success({
+            output: responseResult.value.output,
+            // if there were tool calls, we would have returned already, so we know these are empty
+            toolCalls: [],
+            ...(thinkingBlocks.length > 0 && { thinkingBlocks }),
+            usage: addTokenUsage(toolResult.value.usage, responseResult.value.usage),
+            cost: addCosts(toolResult.value.cost, responseResult.value.cost),
+            model: request.model,
+        });
+    }
+    async __textSync(request) {
         this.logger.debug("Sending request to Google Gemini:", JSON.stringify(request, null, 2));
         // Send the prompt as the latest message
         const result = await this.client.models.generateContent(request);
@@ -136,6 +195,13 @@ export class SmolGoogle extends BaseClient {
         if (signal) {
             request.config = { ...request.config, abortSignal: signal };
         }
+        const hasTools = config.tools && config.tools.length > 0;
+        const hasStructuredResponse = !!config.responseFormat;
+        if (hasTools && hasStructuredResponse) {
+            this.logger.debug("Gemini does not support streaming responses with both tool calls and structured response formats. Response format will be ignored.");
+            request.config.responseMimeType = undefined;
+            request.config.responseJsonSchema = undefined;
+        }
         this.logger.debug("Sending streaming request to Google Gemini:", JSON.stringify(request, null, 2));
         const stream = await this.client.models.generateContentStream(request);
         let content = "";

package/dist/model.js

@@ -109,12 +109,12 @@ export class Model {
         if (!model || !isTextModel(model)) {
             return null;
         }
-        const inputCost = round((usage.inputTokens * (model.inputTokenCost || 0)) / 1_000_000, 2);
-        const outputCost = round((usage.outputTokens * (model.outputTokenCost || 0)) / 1_000_000, 2);
+        const inputCost = round((usage.inputTokens * (model.inputTokenCost || 0)) / 1_000_000, 6);
+        const outputCost = round((usage.outputTokens * (model.outputTokenCost || 0)) / 1_000_000, 6);
         const cachedInputCost = usage.cachedInputTokens && model.cachedInputTokenCost
-            ? round((usage.cachedInputTokens * model.cachedInputTokenCost) / 1_000_000, 2)
+            ? round((usage.cachedInputTokens * model.cachedInputTokenCost) / 1_000_000, 6)
             : undefined;
-        const totalCost = round(inputCost + outputCost + (cachedInputCost || 0), 2);
+        const totalCost = round(inputCost + outputCost + (cachedInputCost || 0), 6);
         return {
             inputCost,
             outputCost,

package/dist/models.d.ts

@@ -24,6 +24,18 @@ export type TextModel = BaseModel & {
     maxInputTokens: number;
     maxOutputTokens: number;
     outputTokensPerSecond?: number;
+    reasoning?: {
+        /** Available effort/thinking levels (provider-specific). Omit for budget-based thinking (Anthropic, Gemini 2.5). */
+        levels?: readonly string[];
+        /** Default reasoning level */
+        defaultLevel?: string;
+        /** Whether reasoning/thinking can be fully disabled */
+        canDisable?: boolean;
+        /** Whether the response includes visible thinking content (thinking blocks/parts) */
+        outputsThinking?: boolean;
+        /** Whether cryptographic thinking signatures are returned for round-tripping */
+        outputsSignatures?: boolean;
+    };
 };
 export type EmbeddingsModel = {
     type: "embeddings";
@@ -73,6 +85,13 @@ export declare const textModels: readonly [{
     readonly cachedInputTokenCost: 0.5;
     readonly outputTokenCost: 8;
     readonly outputTokensPerSecond: 94;
+    readonly reasoning: {
+        readonly levels: readonly ["low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
     readonly provider: "openai";
 }, {
     readonly type: "text";
@@ -84,6 +103,13 @@ export declare const textModels: readonly [{
     readonly cachedInputTokenCost: 0.55;
     readonly outputTokenCost: 4.4;
     readonly outputTokensPerSecond: 214;
+    readonly reasoning: {
+        readonly levels: readonly ["low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
     readonly provider: "openai";
 }, {
     readonly type: "text";
@@ -95,6 +121,13 @@ export declare const textModels: readonly [{
     readonly cachedInputTokenCost: 0.275;
     readonly outputTokenCost: 4.4;
     readonly outputTokensPerSecond: 135;
+    readonly reasoning: {
+        readonly levels: readonly ["low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
     readonly provider: "openai";
 }, {
     readonly type: "text";
@@ -104,6 +137,11 @@ export declare const textModels: readonly [{
     readonly maxOutputTokens: 100000;
     readonly inputTokenCost: 20;
     readonly outputTokenCost: 80;
+    readonly reasoning: {
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
     readonly provider: "openai";
 }, {
     readonly type: "text";
@@ -115,6 +153,13 @@ export declare const textModels: readonly [{
     readonly cachedInputTokenCost: 7.5;
     readonly outputTokenCost: 60;
     readonly outputTokensPerSecond: 100;
+    readonly reasoning: {
+        readonly levels: readonly ["low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
     readonly provider: "openai";
 }, {
     readonly type: "text";
@@ -179,6 +224,128 @@ export declare const textModels: readonly [{
     readonly outputTokenCost: 0.4;
     readonly outputTokensPerSecond: 142;
     readonly provider: "openai";
+}, {
+    readonly type: "text";
+    readonly modelName: "gpt-5";
+    readonly description: "GPT-5 is a frontier reasoning model with 400K context window. Supports reasoning tokens. Knowledge cutoff: September 2024.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 1.25;
+    readonly cachedInputTokenCost: 0.125;
+    readonly outputTokenCost: 10;
+    readonly outputTokensPerSecond: 72;
+    readonly reasoning: {
+        readonly levels: readonly ["minimal", "low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+}, {
+    readonly type: "text";
+    readonly modelName: "gpt-5-mini";
+    readonly description: "GPT-5 mini is a faster, more cost-efficient version of GPT-5 with 400K context window. Knowledge cutoff: May 2024.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 0.25;
+    readonly cachedInputTokenCost: 0.025;
+    readonly outputTokenCost: 2;
+    readonly outputTokensPerSecond: 69;
+    readonly reasoning: {
+        readonly levels: readonly ["minimal", "low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+}, {
+    readonly type: "text";
+    readonly modelName: "gpt-5-nano";
+    readonly description: "GPT-5 nano is the fastest and most affordable GPT-5 variant with 400K context window. Knowledge cutoff: May 2024.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 0.05;
+    readonly cachedInputTokenCost: 0.005;
+    readonly outputTokenCost: 0.4;
+    readonly outputTokensPerSecond: 140;
+    readonly reasoning: {
+        readonly levels: readonly ["minimal", "low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+}, {
+    readonly type: "text";
+    readonly modelName: "gpt-5.1";
+    readonly description: "GPT-5.1 is the flagship model for coding and agentic tasks with configurable reasoning effort. 400K context window. Knowledge cutoff: September 2024.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 1.25;
+    readonly cachedInputTokenCost: 0.125;
+    readonly outputTokenCost: 10;
+    readonly reasoning: {
+        readonly levels: readonly ["none", "low", "medium", "high"];
+        readonly defaultLevel: "none";
+        readonly canDisable: true;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+}, {
+    readonly type: "text";
+    readonly modelName: "gpt-5.2";
+    readonly description: "GPT-5.2 is the flagship model for coding and agentic tasks across industries. 400K context window. Knowledge cutoff: August 2025.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 1.75;
+    readonly cachedInputTokenCost: 0.175;
+    readonly outputTokenCost: 14;
+    readonly outputTokensPerSecond: 61;
+    readonly reasoning: {
+        readonly levels: readonly ["none", "low", "medium", "high"];
+        readonly defaultLevel: "none";
+        readonly canDisable: true;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+}, {
+    readonly type: "text";
+    readonly modelName: "gpt-5.4";
+    readonly description: "GPT-5.4 is the most capable and efficient frontier model for complex professional work. 1M context window, state-of-the-art coding and tool use. Standard pricing for ≤272K tokens, 2x input/1.5x output for >272K. Knowledge cutoff: August 2025.";
+    readonly maxInputTokens: 1050000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 2.5;
+    readonly cachedInputTokenCost: 0.25;
+    readonly outputTokenCost: 15;
+    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";
+    readonly description: "GPT-5.4 Pro uses more compute for complex reasoning tasks. 1M context window. Standard pricing for ≤272K tokens. Knowledge cutoff: August 2025.";
+    readonly maxInputTokens: 1050000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 30;
+    readonly outputTokenCost: 180;
+    readonly reasoning: {
+        readonly levels: readonly ["medium", "high", "xhigh"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
 }, {
     readonly type: "text";
     readonly modelName: "gemini-3.1-pro-preview";
@@ -188,15 +355,23 @@ export declare const textModels: readonly [{
     readonly inputTokenCost: 2;
     readonly outputTokenCost: 12;
     readonly outputTokensPerSecond: 112;
+    readonly reasoning: {
+        readonly levels: readonly ["low", "medium", "high"];
+        readonly defaultLevel: "high";
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "google";
 }, {
     readonly type: "text";
     readonly modelName: "gemini-3-pro-preview";
-    readonly description: "Gemini 3 Pro with 1M context window and 64K output. Standard pricing for ≤200k tokens ($2.00 input/$12.00 output), higher rates for >200k tokens ($4.00 input/$18.00 output). Superseded by 3.1 Pro.";
+    readonly description: "DEPRECATED: Shut down March 9, 2026. Use gemini-3.1-pro-preview instead.";
     readonly maxInputTokens: 1048576;
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 2;
     readonly outputTokenCost: 12;
+    readonly disabled: true;
     readonly provider: "google";
 }, {
     readonly type: "text";
@@ -206,6 +381,31 @@ export declare const textModels: readonly [{
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 0.5;
     readonly outputTokenCost: 3;
+    readonly outputTokensPerSecond: 146;
+    readonly reasoning: {
+        readonly levels: readonly ["minimal", "low", "medium", "high"];
+        readonly defaultLevel: "high";
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
+    readonly provider: "google";
+}, {
+    readonly type: "text";
+    readonly modelName: "gemini-3.1-flash-lite-preview";
+    readonly description: "Most cost-effective Gemini 3.1 model with thinking support and 1M context window. 2.5x faster TTFA and 45% faster output than 2.5 Flash. Released March 2026.";
+    readonly maxInputTokens: 1048576;
+    readonly maxOutputTokens: 65536;
+    readonly inputTokenCost: 0.25;
+    readonly outputTokenCost: 1.5;
+    readonly outputTokensPerSecond: 379;
+    readonly reasoning: {
+        readonly levels: readonly ["minimal", "low", "medium", "high"];
+        readonly defaultLevel: "minimal";
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "google";
 }, {
     readonly type: "text";
@@ -215,7 +415,12 @@ export declare const textModels: readonly [{
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 1.25;
     readonly outputTokenCost: 10;
-    readonly outputTokensPerSecond: 175;
+    readonly outputTokensPerSecond: 134;
+    readonly reasoning: {
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "google";
 }, {
     readonly type: "text";
@@ -225,7 +430,12 @@ export declare const textModels: readonly [{
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 0.3;
     readonly outputTokenCost: 2.5;
-    readonly outputTokensPerSecond: 225;
+    readonly outputTokensPerSecond: 245;
+    readonly reasoning: {
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "google";
 }, {
     readonly type: "text";
@@ -236,6 +446,11 @@ export declare const textModels: readonly [{
     readonly inputTokenCost: 0.1;
     readonly outputTokenCost: 0.4;
     readonly outputTokensPerSecond: 400;
+    readonly reasoning: {
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: false;
+    };
     readonly provider: "google";
 }, {
     readonly type: "text";
@@ -306,20 +521,33 @@ export declare const textModels: readonly [{
 }, {
     readonly type: "text";
     readonly modelName: "claude-opus-4-6";
-    readonly description: "The most intelligent Claude model for building agents and coding. 200K context window, 128K max output.";
+    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 inputTokenCost: 5;
+    readonly cachedInputTokenCost: 0.5;
     readonly outputTokenCost: 25;
+    readonly outputTokensPerSecond: 53;
+    readonly reasoning: {
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "anthropic";
 }, {
     readonly type: "text";
     readonly modelName: "claude-sonnet-4-6";
-    readonly description: "The best combination of speed and intelligence. 200K context window, 64K max output.";
+    readonly description: "The best combination of speed and intelligence. 200K context window (1M in beta), 64K max output.";
     readonly maxInputTokens: 200000;
     readonly maxOutputTokens: 64000;
     readonly inputTokenCost: 3;
+    readonly cachedInputTokenCost: 0.3;
     readonly outputTokenCost: 15;
+    readonly reasoning: {
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "anthropic";
 }, {
     readonly type: "text";
@@ -328,7 +556,14 @@ export declare const textModels: readonly [{
     readonly maxInputTokens: 200000;
     readonly maxOutputTokens: 64000;
     readonly inputTokenCost: 1;
+    readonly cachedInputTokenCost: 0.1;
     readonly outputTokenCost: 5;
+    readonly outputTokensPerSecond: 97;
+    readonly reasoning: {
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "anthropic";
 }, {
     readonly type: "text";
@@ -339,6 +574,11 @@ export declare const textModels: readonly [{
     readonly inputTokenCost: 3;
     readonly outputTokenCost: 15;
     readonly outputTokensPerSecond: 78;
+    readonly reasoning: {
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly disabled: true;
     readonly provider: "anthropic";
 }, {
@@ -465,6 +705,13 @@ export declare function getModel(modelName: ModelName): {
     readonly cachedInputTokenCost: 0.5;
     readonly outputTokenCost: 8;
     readonly outputTokensPerSecond: 94;
+    readonly reasoning: {
+        readonly levels: readonly ["low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
     readonly provider: "openai";
 } | {
     readonly type: "text";
@@ -476,6 +723,13 @@ export declare function getModel(modelName: ModelName): {
     readonly cachedInputTokenCost: 0.55;
     readonly outputTokenCost: 4.4;
     readonly outputTokensPerSecond: 214;
+    readonly reasoning: {
+        readonly levels: readonly ["low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
     readonly provider: "openai";
 } | {
     readonly type: "text";
@@ -487,6 +741,13 @@ export declare function getModel(modelName: ModelName): {
     readonly cachedInputTokenCost: 0.275;
     readonly outputTokenCost: 4.4;
     readonly outputTokensPerSecond: 135;
+    readonly reasoning: {
+        readonly levels: readonly ["low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
     readonly provider: "openai";
 } | {
     readonly type: "text";
@@ -496,6 +757,11 @@ export declare function getModel(modelName: ModelName): {
     readonly maxOutputTokens: 100000;
     readonly inputTokenCost: 20;
     readonly outputTokenCost: 80;
+    readonly reasoning: {
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
     readonly provider: "openai";
 } | {
     readonly type: "text";
@@ -507,6 +773,13 @@ export declare function getModel(modelName: ModelName): {
     readonly cachedInputTokenCost: 7.5;
     readonly outputTokenCost: 60;
     readonly outputTokensPerSecond: 100;
+    readonly reasoning: {
+        readonly levels: readonly ["low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
     readonly provider: "openai";
 } | {
     readonly type: "text";
@@ -571,6 +844,128 @@ export declare function getModel(modelName: ModelName): {
     readonly outputTokenCost: 0.4;
     readonly outputTokensPerSecond: 142;
     readonly provider: "openai";
+} | {
+    readonly type: "text";
+    readonly modelName: "gpt-5";
+    readonly description: "GPT-5 is a frontier reasoning model with 400K context window. Supports reasoning tokens. Knowledge cutoff: September 2024.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 1.25;
+    readonly cachedInputTokenCost: 0.125;
+    readonly outputTokenCost: 10;
+    readonly outputTokensPerSecond: 72;
+    readonly reasoning: {
+        readonly levels: readonly ["minimal", "low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+} | {
+    readonly type: "text";
+    readonly modelName: "gpt-5-mini";
+    readonly description: "GPT-5 mini is a faster, more cost-efficient version of GPT-5 with 400K context window. Knowledge cutoff: May 2024.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 0.25;
+    readonly cachedInputTokenCost: 0.025;
+    readonly outputTokenCost: 2;
+    readonly outputTokensPerSecond: 69;
+    readonly reasoning: {
+        readonly levels: readonly ["minimal", "low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+} | {
+    readonly type: "text";
+    readonly modelName: "gpt-5-nano";
+    readonly description: "GPT-5 nano is the fastest and most affordable GPT-5 variant with 400K context window. Knowledge cutoff: May 2024.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 0.05;
+    readonly cachedInputTokenCost: 0.005;
+    readonly outputTokenCost: 0.4;
+    readonly outputTokensPerSecond: 140;
+    readonly reasoning: {
+        readonly levels: readonly ["minimal", "low", "medium", "high"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+} | {
+    readonly type: "text";
+    readonly modelName: "gpt-5.1";
+    readonly description: "GPT-5.1 is the flagship model for coding and agentic tasks with configurable reasoning effort. 400K context window. Knowledge cutoff: September 2024.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 1.25;
+    readonly cachedInputTokenCost: 0.125;
+    readonly outputTokenCost: 10;
+    readonly reasoning: {
+        readonly levels: readonly ["none", "low", "medium", "high"];
+        readonly defaultLevel: "none";
+        readonly canDisable: true;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+} | {
+    readonly type: "text";
+    readonly modelName: "gpt-5.2";
+    readonly description: "GPT-5.2 is the flagship model for coding and agentic tasks across industries. 400K context window. Knowledge cutoff: August 2025.";
+    readonly maxInputTokens: 400000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 1.75;
+    readonly cachedInputTokenCost: 0.175;
+    readonly outputTokenCost: 14;
+    readonly outputTokensPerSecond: 61;
+    readonly reasoning: {
+        readonly levels: readonly ["none", "low", "medium", "high"];
+        readonly defaultLevel: "none";
+        readonly canDisable: true;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
+} | {
+    readonly type: "text";
+    readonly modelName: "gpt-5.4";
+    readonly description: "GPT-5.4 is the most capable and efficient frontier model for complex professional work. 1M context window, state-of-the-art coding and tool use. Standard pricing for ≤272K tokens, 2x input/1.5x output for >272K. Knowledge cutoff: August 2025.";
+    readonly maxInputTokens: 1050000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 2.5;
+    readonly cachedInputTokenCost: 0.25;
+    readonly outputTokenCost: 15;
+    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";
+    readonly description: "GPT-5.4 Pro uses more compute for complex reasoning tasks. 1M context window. Standard pricing for ≤272K tokens. Knowledge cutoff: August 2025.";
+    readonly maxInputTokens: 1050000;
+    readonly maxOutputTokens: 128000;
+    readonly inputTokenCost: 30;
+    readonly outputTokenCost: 180;
+    readonly reasoning: {
+        readonly levels: readonly ["medium", "high", "xhigh"];
+        readonly defaultLevel: "medium";
+        readonly canDisable: false;
+        readonly outputsThinking: false;
+        readonly outputsSignatures: false;
+    };
+    readonly provider: "openai";
 } | {
     readonly type: "text";
     readonly modelName: "gemini-3.1-pro-preview";
@@ -580,15 +975,23 @@ export declare function getModel(modelName: ModelName): {
     readonly inputTokenCost: 2;
     readonly outputTokenCost: 12;
     readonly outputTokensPerSecond: 112;
+    readonly reasoning: {
+        readonly levels: readonly ["low", "medium", "high"];
+        readonly defaultLevel: "high";
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "google";
 } | {
     readonly type: "text";
     readonly modelName: "gemini-3-pro-preview";
-    readonly description: "Gemini 3 Pro with 1M context window and 64K output. Standard pricing for ≤200k tokens ($2.00 input/$12.00 output), higher rates for >200k tokens ($4.00 input/$18.00 output). Superseded by 3.1 Pro.";
+    readonly description: "DEPRECATED: Shut down March 9, 2026. Use gemini-3.1-pro-preview instead.";
     readonly maxInputTokens: 1048576;
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 2;
     readonly outputTokenCost: 12;
+    readonly disabled: true;
     readonly provider: "google";
 } | {
     readonly type: "text";
@@ -598,6 +1001,31 @@ export declare function getModel(modelName: ModelName): {
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 0.5;
     readonly outputTokenCost: 3;
+    readonly outputTokensPerSecond: 146;
+    readonly reasoning: {
+        readonly levels: readonly ["minimal", "low", "medium", "high"];
+        readonly defaultLevel: "high";
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
+    readonly provider: "google";
+} | {
+    readonly type: "text";
+    readonly modelName: "gemini-3.1-flash-lite-preview";
+    readonly description: "Most cost-effective Gemini 3.1 model with thinking support and 1M context window. 2.5x faster TTFA and 45% faster output than 2.5 Flash. Released March 2026.";
+    readonly maxInputTokens: 1048576;
+    readonly maxOutputTokens: 65536;
+    readonly inputTokenCost: 0.25;
+    readonly outputTokenCost: 1.5;
+    readonly outputTokensPerSecond: 379;
+    readonly reasoning: {
+        readonly levels: readonly ["minimal", "low", "medium", "high"];
+        readonly defaultLevel: "minimal";
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "google";
 } | {
     readonly type: "text";
@@ -607,7 +1035,12 @@ export declare function getModel(modelName: ModelName): {
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 1.25;
     readonly outputTokenCost: 10;
-    readonly outputTokensPerSecond: 175;
+    readonly outputTokensPerSecond: 134;
+    readonly reasoning: {
+        readonly canDisable: false;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "google";
 } | {
     readonly type: "text";
@@ -617,7 +1050,12 @@ export declare function getModel(modelName: ModelName): {
     readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 0.3;
     readonly outputTokenCost: 2.5;
-    readonly outputTokensPerSecond: 225;
+    readonly outputTokensPerSecond: 245;
+    readonly reasoning: {
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "google";
 } | {
     readonly type: "text";
@@ -628,6 +1066,11 @@ export declare function getModel(modelName: ModelName): {
     readonly inputTokenCost: 0.1;
     readonly outputTokenCost: 0.4;
     readonly outputTokensPerSecond: 400;
+    readonly reasoning: {
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: false;
+    };
     readonly provider: "google";
 } | {
     readonly type: "text";
@@ -698,20 +1141,33 @@ export declare function getModel(modelName: ModelName): {
 } | {
     readonly type: "text";
     readonly modelName: "claude-opus-4-6";
-    readonly description: "The most intelligent Claude model for building agents and coding. 200K context window, 128K max output.";
+    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 inputTokenCost: 5;
+    readonly cachedInputTokenCost: 0.5;
     readonly outputTokenCost: 25;
+    readonly outputTokensPerSecond: 53;
+    readonly reasoning: {
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "anthropic";
 } | {
     readonly type: "text";
     readonly modelName: "claude-sonnet-4-6";
-    readonly description: "The best combination of speed and intelligence. 200K context window, 64K max output.";
+    readonly description: "The best combination of speed and intelligence. 200K context window (1M in beta), 64K max output.";
     readonly maxInputTokens: 200000;
     readonly maxOutputTokens: 64000;
     readonly inputTokenCost: 3;
+    readonly cachedInputTokenCost: 0.3;
     readonly outputTokenCost: 15;
+    readonly reasoning: {
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "anthropic";
 } | {
     readonly type: "text";
@@ -720,7 +1176,14 @@ export declare function getModel(modelName: ModelName): {
     readonly maxInputTokens: 200000;
     readonly maxOutputTokens: 64000;
     readonly inputTokenCost: 1;
+    readonly cachedInputTokenCost: 0.1;
     readonly outputTokenCost: 5;
+    readonly outputTokensPerSecond: 97;
+    readonly reasoning: {
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly provider: "anthropic";
 } | {
     readonly type: "text";
@@ -731,6 +1194,11 @@ export declare function getModel(modelName: ModelName): {
     readonly inputTokenCost: 3;
     readonly outputTokenCost: 15;
     readonly outputTokensPerSecond: 78;
+    readonly reasoning: {
+        readonly canDisable: true;
+        readonly outputsThinking: true;
+        readonly outputsSignatures: true;
+    };
     readonly disabled: true;
     readonly provider: "anthropic";
 } | {

package/dist/models.js

@@ -52,6 +52,13 @@ export const textModels = [
         cachedInputTokenCost: 0.5,
         outputTokenCost: 8,
         outputTokensPerSecond: 94,
+        reasoning: {
+            levels: ["low", "medium", "high"],
+            defaultLevel: "medium",
+            canDisable: false,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
         provider: "openai",
     },
     {
@@ -64,6 +71,13 @@ export const textModels = [
         cachedInputTokenCost: 0.55,
         outputTokenCost: 4.4,
         outputTokensPerSecond: 214,
+        reasoning: {
+            levels: ["low", "medium", "high"],
+            defaultLevel: "medium",
+            canDisable: false,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
         provider: "openai",
     },
     {
@@ -76,6 +90,13 @@ export const textModels = [
         cachedInputTokenCost: 0.275,
         outputTokenCost: 4.4,
         outputTokensPerSecond: 135,
+        reasoning: {
+            levels: ["low", "medium", "high"],
+            defaultLevel: "medium",
+            canDisable: false,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
         provider: "openai",
     },
     {
@@ -86,6 +107,11 @@ export const textModels = [
         maxOutputTokens: 100000,
         inputTokenCost: 20,
         outputTokenCost: 80,
+        reasoning: {
+            canDisable: false,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
         provider: "openai",
     },
     {
@@ -98,6 +124,13 @@ export const textModels = [
         cachedInputTokenCost: 7.5,
         outputTokenCost: 60,
         outputTokensPerSecond: 100,
+        reasoning: {
+            levels: ["low", "medium", "high"],
+            defaultLevel: "medium",
+            canDisable: false,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
         provider: "openai",
     },
     {
@@ -169,6 +202,135 @@ export const textModels = [
         outputTokensPerSecond: 142,
         provider: "openai",
     },
+    {
+        type: "text",
+        modelName: "gpt-5",
+        description: "GPT-5 is a frontier reasoning model with 400K context window. Supports reasoning tokens. Knowledge cutoff: September 2024.",
+        maxInputTokens: 400000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 1.25,
+        cachedInputTokenCost: 0.125,
+        outputTokenCost: 10,
+        outputTokensPerSecond: 72,
+        reasoning: {
+            levels: ["minimal", "low", "medium", "high"],
+            defaultLevel: "medium",
+            canDisable: false,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        provider: "openai",
+    },
+    {
+        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.",
+        maxInputTokens: 400000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 0.25,
+        cachedInputTokenCost: 0.025,
+        outputTokenCost: 2,
+        outputTokensPerSecond: 69,
+        reasoning: {
+            levels: ["minimal", "low", "medium", "high"],
+            defaultLevel: "medium",
+            canDisable: false,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        provider: "openai",
+    },
+    {
+        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.",
+        maxInputTokens: 400000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 0.05,
+        cachedInputTokenCost: 0.005,
+        outputTokenCost: 0.4,
+        outputTokensPerSecond: 140,
+        reasoning: {
+            levels: ["minimal", "low", "medium", "high"],
+            defaultLevel: "medium",
+            canDisable: false,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        provider: "openai",
+    },
+    {
+        type: "text",
+        modelName: "gpt-5.1",
+        description: "GPT-5.1 is the flagship model for coding and agentic tasks with configurable reasoning effort. 400K context window. Knowledge cutoff: September 2024.",
+        maxInputTokens: 400000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 1.25,
+        cachedInputTokenCost: 0.125,
+        outputTokenCost: 10,
+        reasoning: {
+            levels: ["none", "low", "medium", "high"],
+            defaultLevel: "none",
+            canDisable: true,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        provider: "openai",
+    },
+    {
+        type: "text",
+        modelName: "gpt-5.2",
+        description: "GPT-5.2 is the flagship model for coding and agentic tasks across industries. 400K context window. Knowledge cutoff: August 2025.",
+        maxInputTokens: 400000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 1.75,
+        cachedInputTokenCost: 0.175,
+        outputTokenCost: 14,
+        outputTokensPerSecond: 61,
+        reasoning: {
+            levels: ["none", "low", "medium", "high"],
+            defaultLevel: "none",
+            canDisable: true,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        provider: "openai",
+    },
+    {
+        type: "text",
+        modelName: "gpt-5.4",
+        description: "GPT-5.4 is the most capable and efficient frontier model for complex professional work. 1M context window, state-of-the-art coding and tool use. Standard pricing for ≤272K tokens, 2x input/1.5x output for >272K. Knowledge cutoff: August 2025.",
+        maxInputTokens: 1_050_000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 2.5,
+        cachedInputTokenCost: 0.25,
+        outputTokenCost: 15,
+        reasoning: {
+            levels: ["none", "low", "medium", "high", "xhigh"],
+            defaultLevel: "none",
+            canDisable: true,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        provider: "openai",
+    },
+    {
+        type: "text",
+        modelName: "gpt-5.4-pro",
+        description: "GPT-5.4 Pro uses more compute for complex reasoning tasks. 1M context window. Standard pricing for ≤272K tokens. Knowledge cutoff: August 2025.",
+        maxInputTokens: 1_050_000,
+        maxOutputTokens: 128000,
+        inputTokenCost: 30,
+        outputTokenCost: 180,
+        reasoning: {
+            levels: ["medium", "high", "xhigh"],
+            defaultLevel: "medium",
+            canDisable: false,
+            outputsThinking: false,
+            outputsSignatures: false,
+        },
+        provider: "openai",
+    },
     {
         type: "text",
         modelName: "gemini-3.1-pro-preview",
@@ -178,16 +340,24 @@ export const textModels = [
         inputTokenCost: 2.0,
         outputTokenCost: 12.0,
         outputTokensPerSecond: 112,
+        reasoning: {
+            levels: ["low", "medium", "high"],
+            defaultLevel: "high",
+            canDisable: false,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
         provider: "google",
     },
     {
         type: "text",
         modelName: "gemini-3-pro-preview",
-        description: "Gemini 3 Pro with 1M context window and 64K output. Standard pricing for ≤200k tokens ($2.00 input/$12.00 output), higher rates for >200k tokens ($4.00 input/$18.00 output). Superseded by 3.1 Pro.",
+        description: "DEPRECATED: Shut down March 9, 2026. Use gemini-3.1-pro-preview instead.",
         maxInputTokens: 1_048_576,
         maxOutputTokens: 65536,
         inputTokenCost: 2.0,
         outputTokenCost: 12.0,
+        disabled: true,
         provider: "google",
     },
     {
@@ -198,6 +368,32 @@ export const textModels = [
         maxOutputTokens: 65536,
         inputTokenCost: 0.5,
         outputTokenCost: 3.0,
+        outputTokensPerSecond: 146,
+        reasoning: {
+            levels: ["minimal", "low", "medium", "high"],
+            defaultLevel: "high",
+            canDisable: false,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
+        provider: "google",
+    },
+    {
+        type: "text",
+        modelName: "gemini-3.1-flash-lite-preview",
+        description: "Most cost-effective Gemini 3.1 model with thinking support and 1M context window. 2.5x faster TTFA and 45% faster output than 2.5 Flash. Released March 2026.",
+        maxInputTokens: 1_048_576,
+        maxOutputTokens: 65536,
+        inputTokenCost: 0.25,
+        outputTokenCost: 1.5,
+        outputTokensPerSecond: 379,
+        reasoning: {
+            levels: ["minimal", "low", "medium", "high"],
+            defaultLevel: "minimal",
+            canDisable: false,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
         provider: "google",
     },
     {
@@ -208,7 +404,12 @@ export const textModels = [
         maxOutputTokens: 65536,
         inputTokenCost: 1.25,
         outputTokenCost: 10.0,
-        outputTokensPerSecond: 175,
+        outputTokensPerSecond: 134,
+        reasoning: {
+            canDisable: false,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
         provider: "google",
     },
     {
@@ -219,7 +420,12 @@ export const textModels = [
         maxOutputTokens: 65536,
         inputTokenCost: 0.3,
         outputTokenCost: 2.5,
-        outputTokensPerSecond: 225,
+        outputTokensPerSecond: 245,
+        reasoning: {
+            canDisable: true,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
         provider: "google",
     },
     {
@@ -231,6 +437,11 @@ export const textModels = [
         inputTokenCost: 0.1,
         outputTokenCost: 0.4,
         outputTokensPerSecond: 400,
+        reasoning: {
+            canDisable: true,
+            outputsThinking: true,
+            outputsSignatures: false,
+        },
         provider: "google",
     },
     {
@@ -308,21 +519,34 @@ export const textModels = [
     {
         type: "text",
         modelName: "claude-opus-4-6",
-        description: "The most intelligent Claude model for building agents and coding. 200K context window, 128K max output.",
+        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,
         inputTokenCost: 5,
+        cachedInputTokenCost: 0.5,
         outputTokenCost: 25,
+        outputTokensPerSecond: 53,
+        reasoning: {
+            canDisable: true,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
         provider: "anthropic",
     },
     {
         type: "text",
         modelName: "claude-sonnet-4-6",
-        description: "The best combination of speed and intelligence. 200K context window, 64K max output.",
+        description: "The best combination of speed and intelligence. 200K context window (1M in beta), 64K max output.",
         maxInputTokens: 200_000,
         maxOutputTokens: 64_000,
         inputTokenCost: 3,
+        cachedInputTokenCost: 0.3,
         outputTokenCost: 15,
+        reasoning: {
+            canDisable: true,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
         provider: "anthropic",
     },
     {
@@ -332,7 +556,14 @@ export const textModels = [
         maxInputTokens: 200_000,
         maxOutputTokens: 64_000,
         inputTokenCost: 1,
+        cachedInputTokenCost: 0.1,
         outputTokenCost: 5,
+        outputTokensPerSecond: 97,
+        reasoning: {
+            canDisable: true,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
         provider: "anthropic",
     },
     {
@@ -344,6 +575,11 @@ export const textModels = [
         inputTokenCost: 3,
         outputTokenCost: 15,
         outputTokensPerSecond: 78,
+        reasoning: {
+            canDisable: true,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
         disabled: true,
         provider: "anthropic",
     },

package/dist/strategies/raceStrategy.js

@@ -1,3 +1,4 @@
+import { getLogger } from "../logger.js";
 import { BaseStrategy } from "./baseStrategy.js";
 export class RaceStrategy extends BaseStrategy {
     strategies;
@@ -28,8 +29,11 @@ export class RaceStrategy extends BaseStrategy {
         });
         return Promise.race(promises.map((p, i) => p.then((result) => {
             for (let j = 0; j < controllers.length; j++) {
-                if (j !== i)
+                if (j !== i) {
+                    const logger = getLogger();
+                    logger.debug(`RaceStrategy: aborting strategy ${this.strategies[j]} because strategy ${this.strategies[i]} won the race.`);
                     controllers[j].abort();
+                }
             }
             return result;
         })));

package/dist/types.d.ts

@@ -98,6 +98,8 @@ export type CostEstimate = {
     totalCost: number;
     currency: string;
 };
+export declare function addTokenUsage(_a?: TokenUsage, _b?: TokenUsage): TokenUsage;
+export declare function addCosts(_a?: CostEstimate, _b?: CostEstimate): CostEstimate;
 export type PromptResult = {
     output: string | null;
     toolCalls: ToolCall[];

package/dist/types.js

@@ -1 +1,41 @@
 export * from "./types/result.js";
+export function addTokenUsage(_a, _b) {
+    let a = _a;
+    let b = _b;
+    if (a && !b)
+        return a;
+    if (b && !a)
+        return b;
+    if (!a && !b)
+        return { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
+    a = _a;
+    b = _b;
+    return {
+        inputTokens: a.inputTokens + b.inputTokens,
+        outputTokens: a.outputTokens + b.outputTokens,
+        cachedInputTokens: (a.cachedInputTokens || 0) + (b.cachedInputTokens || 0),
+        totalTokens: (a.totalTokens || 0) + (b.totalTokens || 0),
+    };
+}
+export function addCosts(_a, _b) {
+    let a = _a;
+    let b = _b;
+    if (a && !b)
+        return a;
+    if (b && !a)
+        return b;
+    if (!a && !b)
+        return { inputCost: 0, outputCost: 0, totalCost: 0, currency: "USD" };
+    a = _a;
+    b = _b;
+    if (a.currency !== b.currency) {
+        throw new Error(`Cannot add costs with different currencies: ${a.currency} and ${b.currency}`);
+    }
+    return {
+        inputCost: a.inputCost + b.inputCost,
+        outputCost: a.outputCost + b.outputCost,
+        cachedInputCost: (a.cachedInputCost || 0) + (b.cachedInputCost || 0),
+        totalCost: a.totalCost + b.totalCost,
+        currency: a.currency,
+    };
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "smoltalk",
-  "version": "0.0.43",
+  "version": "0.0.45",
   "description": "A common interface for LLM APIs",
   "homepage": "https://github.com/egonSchiele/smoltalk",
   "scripts": {