smoltalk 0.0.16 → 0.0.18

package/dist/classes/message/AssistantMessage.d.ts

@@ -20,6 +20,7 @@ export declare class AssistantMessage extends BaseMessage implements MessageClas
         rawData?: any;
     });
     get content(): string;
+    set content(value: string);
     get role(): "assistant";
     get name(): string | undefined;
     get audio(): any | null | undefined;

package/dist/classes/message/AssistantMessage.js

@@ -25,6 +25,9 @@ export class AssistantMessage extends BaseMessage {
             ? this._content
             : JSON.stringify(this._content);
     }
+    set content(value) {
+        this._content = value;
+    }
     get role() {
         return this._role;
     }

package/dist/classes/message/DeveloperMessage.d.ts

@@ -13,6 +13,7 @@ export declare class DeveloperMessage extends BaseMessage implements MessageClas
         rawData?: any;
     });
     get content(): string;
+    set content(value: string);
     get role(): "developer";
     get name(): string | undefined;
     get rawData(): any;

package/dist/classes/message/DeveloperMessage.js

@@ -15,6 +15,9 @@ export class DeveloperMessage extends BaseMessage {
             ? this._content
             : JSON.stringify(this._content);
     }
+    set content(value) {
+        this._content = value;
+    }
     get role() {
         return this._role;
     }

package/dist/classes/message/SystemMessage.d.ts

@@ -13,6 +13,7 @@ export declare class SystemMessage extends BaseMessage implements MessageClass {
         rawData?: any;
     });
     get content(): string;
+    set content(value: string);
     get role(): "system";
     get name(): string | undefined;
     get rawData(): any;

package/dist/classes/message/SystemMessage.js

@@ -15,6 +15,9 @@ export class SystemMessage extends BaseMessage {
             ? this._content
             : JSON.stringify(this._content);
     }
+    set content(value) {
+        this._content = value;
+    }
     get role() {
         return this._role;
     }

package/dist/classes/message/ToolMessage.d.ts

@@ -15,6 +15,7 @@ export declare class ToolMessage extends BaseMessage implements MessageClass {
         name: string;
     });
     get content(): string;
+    set content(value: string);
     get role(): "tool";
     get name(): string;
     get tool_call_id(): string;

package/dist/classes/message/ToolMessage.js

@@ -17,6 +17,9 @@ export class ToolMessage extends BaseMessage {
             ? this._content
             : JSON.stringify(this._content);
     }
+    set content(value) {
+        this._content = value;
+    }
     get role() {
         return this._role;
     }

package/dist/classes/message/UserMessage.d.ts

@@ -12,6 +12,7 @@ export declare class UserMessage extends BaseMessage implements MessageClass {
         rawData?: any;
     });
     get content(): string;
+    set content(value: string);
     get role(): "user";
     get name(): string | undefined;
     get rawData(): any;

package/dist/classes/message/UserMessage.js

@@ -13,6 +13,9 @@ export class UserMessage extends BaseMessage {
     get content() {
         return this._content;
     }
+    set content(value) {
+        this._content = value;
+    }
     get role() {
         return this._role;
     }

package/dist/clients/google.d.ts

@@ -1,5 +1,5 @@
 import { GoogleGenAI } from "@google/genai";
-import { BaseClientConfig, PromptConfig, PromptResult, Result, SmolClient } from "../types.js";
+import { BaseClientConfig, PromptConfig, PromptResult, Result, SmolClient, StreamChunk } from "../types.js";
 import { BaseClient } from "./baseClient.js";
 export type SmolGoogleConfig = BaseClientConfig;
 export declare class SmolGoogle extends BaseClient implements SmolClient {
@@ -9,5 +9,8 @@ export declare class SmolGoogle extends BaseClient implements SmolClient {
     constructor(config: SmolGoogleConfig);
     getClient(): GoogleGenAI;
     getModel(): string;
+    private calculateUsageAndCost;
+    private buildRequest;
     _textSync(config: PromptConfig): Promise<Result<PromptResult>>;
+    _textStream(config: PromptConfig): AsyncGenerator<StreamChunk>;
 }

package/dist/clients/google.js

@@ -4,6 +4,7 @@ import { getLogger } from "../logger.js";
 import { success, } from "../types.js";
 import { zodToGoogleTool } from "../util/tool.js";
 import { BaseClient } from "./baseClient.js";
+import { calculateCost } from "../models.js";
 export class SmolGoogle extends BaseClient {
     client;
     logger;
@@ -23,7 +24,24 @@ export class SmolGoogle extends BaseClient {
     getModel() {
         return this.model;
     }
-    async _textSync(config) {
+    calculateUsageAndCost(usageMetadata) {
+        let usage;
+        let cost;
+        if (usageMetadata) {
+            usage = {
+                inputTokens: usageMetadata.promptTokenCount || 0,
+                outputTokens: usageMetadata.candidatesTokenCount || 0,
+                cachedInputTokens: usageMetadata.cachedContentTokenCount,
+                totalTokens: usageMetadata.totalTokenCount,
+            };
+            const calculatedCost = calculateCost(this.model, usage);
+            if (calculatedCost) {
+                cost = calculatedCost;
+            }
+        }
+        return { usage, cost };
+    }
+    buildRequest(config) {
         const messages = config.messages.map((msg) => msg.toGoogleMessage());
         const tools = (config.tools || []).map((tool) => {
             return zodToGoogleTool(tool.name, tool.schema, {
@@ -38,13 +56,18 @@ export class SmolGoogle extends BaseClient {
             genConfig.responseMimeType = "application/json";
             genConfig.responseJsonSchema = config.responseFormat.toJSONSchema();
         }
-        const request = {
+        return {
             contents: messages,
             model: this.model,
             config: genConfig,
-            stream: config.stream || false,
             ...(config.rawAttributes || {}),
         };
+    }
+    async _textSync(config) {
+        const request = {
+            ...this.buildRequest(config),
+            stream: config.stream || false,
+        };
         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);
@@ -61,7 +84,57 @@ export class SmolGoogle extends BaseClient {
                 });
             }
         });
+        // Extract usage and calculate cost
+        const { usage, cost } = this.calculateUsageAndCost(result.usageMetadata);
         // Return the response, updating the chat history
-        return success({ output, toolCalls });
+        return success({ output, toolCalls, usage, cost });
+    }
+    async *_textStream(config) {
+        const request = this.buildRequest(config);
+        this.logger.debug("Sending streaming request to Google Gemini:", JSON.stringify(request, null, 2));
+        const stream = await this.client.models.generateContentStream(request);
+        let content = "";
+        const toolCallsMap = new Map();
+        let usage;
+        let cost;
+        for await (const chunk of stream) {
+            // Extract usage metadata from chunks
+            if (chunk.usageMetadata) {
+                const usageAndCost = this.calculateUsageAndCost(chunk.usageMetadata);
+                usage = usageAndCost.usage;
+                cost = usageAndCost.cost;
+            }
+            // Handle text content
+            if (chunk.text) {
+                content += chunk.text;
+                yield { type: "text", text: chunk.text };
+            }
+            // Handle function calls
+            if (chunk.functionCalls) {
+                for (const functionCall of chunk.functionCalls) {
+                    const id = functionCall.id || functionCall.name || "";
+                    const name = functionCall.name || "";
+                    if (!toolCallsMap.has(id)) {
+                        toolCallsMap.set(id, {
+                            id: id,
+                            name: name,
+                            arguments: functionCall.args,
+                        });
+                    }
+                }
+            }
+        }
+        this.logger.debug("Streaming response completed from Google Gemini");
+        // Yield tool calls
+        const toolCalls = [];
+        for (const tc of toolCallsMap.values()) {
+            const toolCall = new ToolCall(tc.id, tc.name, tc.arguments);
+            toolCalls.push(toolCall);
+            yield { type: "tool_call", toolCall };
+        }
+        yield {
+            type: "done",
+            result: { output: content || null, toolCalls, usage, cost },
+        };
     }
 }

package/dist/clients/ollama.d.ts

@@ -1,5 +1,5 @@
 import { Ollama } from "ollama";
-import { BaseClientConfig, PromptConfig, PromptResult, Result, SmolClient } from "../types.js";
+import { BaseClientConfig, PromptConfig, PromptResult, Result, SmolClient, StreamChunk } from "../types.js";
 import { BaseClient } from "./baseClient.js";
 export declare const DEFAULT_OLLAMA_HOST = "http://localhost:11434";
 export type SmolOllamaConfig = BaseClientConfig;
@@ -10,5 +10,7 @@ export declare class SmolOllama extends BaseClient implements SmolClient {
     constructor(config: SmolOllamaConfig);
     getClient(): Ollama;
     getModel(): string;
+    private calculateUsageAndCost;
     _textSync(config: PromptConfig): Promise<Result<PromptResult>>;
+    _textStream(config: PromptConfig): AsyncGenerator<StreamChunk>;
 }

package/dist/clients/ollama.js

@@ -4,6 +4,7 @@ import { getLogger } from "../logger.js";
 import { success, } from "../types.js";
 import { zodToGoogleTool } from "../util/tool.js";
 import { BaseClient } from "./baseClient.js";
+import { calculateCost } from "../models.js";
 export const DEFAULT_OLLAMA_HOST = "http://localhost:11434";
 export class SmolOllama extends BaseClient {
     logger;
@@ -30,6 +31,24 @@ export class SmolOllama extends BaseClient {
     getModel() {
         return this.model;
     }
+    calculateUsageAndCost(responseData) {
+        let usage;
+        let cost;
+        if (responseData) {
+            const inputTokens = responseData.prompt_eval_count || 0;
+            const outputTokens = responseData.eval_count || 0;
+            usage = {
+                inputTokens,
+                outputTokens,
+                totalTokens: inputTokens + outputTokens,
+            };
+            const calculatedCost = calculateCost(this.model, usage);
+            if (calculatedCost) {
+                cost = calculatedCost;
+            }
+        }
+        return { usage, cost };
+    }
     async _textSync(config) {
         const messages = config.messages.map((msg) => msg.toOpenAIMessage());
         const tools = (config.tools || []).map((tool) => {
@@ -62,7 +81,90 @@ export class SmolOllama extends BaseClient {
                 toolCalls.push(new ToolCall(tool_call.id, tool_call.function.name, tool_call.function.arguments || {}));
             }
         }
+        // Extract usage and calculate cost
+        const { usage, cost } = this.calculateUsageAndCost(result);
         // Return the response, updating the chat history
-        return success({ output, toolCalls });
+        return success({ output, toolCalls, usage, cost });
+    }
+    async *_textStream(config) {
+        const messages = config.messages.map((msg) => msg.toOpenAIMessage());
+        const tools = (config.tools || []).map((tool) => {
+            return zodToGoogleTool(tool.name, tool.schema, {
+                description: tool.description,
+            });
+        });
+        const request = {
+            messages: messages,
+            model: this.model,
+            stream: true,
+        };
+        if (tools.length > 0) {
+            request.tools = tools.map((t) => ({ type: "function", function: t }));
+        }
+        if (config.responseFormat) {
+            request.format = config.responseFormat.toJSONSchema();
+        }
+        if (config.rawAttributes) {
+            Object.assign(request, config.rawAttributes);
+        }
+        this.logger.debug("Sending streaming request to Ollama:", JSON.stringify(request, null, 2));
+        // @ts-ignore
+        const stream = await this.client.chat(request);
+        let content = "";
+        const toolCallsMap = new Map();
+        let usage;
+        let cost;
+        let lastChunk;
+        for await (const chunk of stream) {
+            lastChunk = chunk;
+            // Handle text content
+            if (chunk.message?.content) {
+                content += chunk.message.content;
+                yield { type: "text", text: chunk.message.content };
+            }
+            // Handle tool calls
+            if (chunk.message?.tool_calls) {
+                for (const tc of chunk.message.tool_calls) {
+                    const tool_call = tc;
+                    const id = tool_call.id || tool_call.function.name || "";
+                    const name = tool_call.function.name || "";
+                    if (!toolCallsMap.has(id)) {
+                        toolCallsMap.set(id, {
+                            id: id,
+                            name: name,
+                            arguments: tool_call.function.arguments || {},
+                        });
+                    }
+                    else {
+                        // Merge arguments if tool call is split across chunks
+                        const existing = toolCallsMap.get(id);
+                        if (tool_call.function.arguments) {
+                            existing.arguments = {
+                                ...existing.arguments,
+                                ...tool_call.function.arguments,
+                            };
+                        }
+                    }
+                }
+            }
+        }
+        this.logger.debug("Streaming response completed from Ollama");
+        // Extract usage from the last chunk
+        if (lastChunk) {
+            const usageAndCost = this.calculateUsageAndCost(lastChunk);
+            usage = usageAndCost.usage;
+            cost = usageAndCost.cost;
+        }
+        // Yield tool calls
+        const toolCalls = [];
+        for (const tc of toolCallsMap.values()) {
+            const toolCall = new ToolCall(tc.id, tc.name, tc.arguments);
+            toolCalls.push(toolCall);
+            yield { type: "tool_call", toolCall };
+        }
+        yield {
+            type: "done",
+            result: { output: content || null, toolCalls, usage, cost },
+        };
     }
 }

package/dist/clients/openai.d.ts

@@ -9,6 +9,7 @@ export declare class SmolOpenAi extends BaseClient implements SmolClient {
     constructor(config: SmolOpenAiConfig);
     getClient(): OpenAI;
     getModel(): string;
+    private calculateUsageAndCost;
     private buildRequest;
     _textSync(config: PromptConfig): Promise<Result<PromptResult>>;
     _textStream(config: PromptConfig): AsyncGenerator<StreamChunk>;

package/dist/clients/openai.js

@@ -5,6 +5,7 @@ import { isFunctionToolCall } from "../util.js";
 import { getLogger } from "../logger.js";
 import { BaseClient } from "./baseClient.js";
 import { zodToOpenAITool } from "../util/tool.js";
+import { calculateCost } from "../models.js";
 export class SmolOpenAi extends BaseClient {
     client;
     logger;
@@ -24,6 +25,23 @@ export class SmolOpenAi extends BaseClient {
     getModel() {
         return this.model;
     }
+    calculateUsageAndCost(usageData) {
+        let usage;
+        let cost;
+        if (usageData) {
+            usage = {
+                inputTokens: usageData.prompt_tokens || 0,
+                outputTokens: usageData.completion_tokens || 0,
+                cachedInputTokens: usageData.prompt_tokens_details?.cached_tokens,
+                totalTokens: usageData.total_tokens,
+            };
+            const calculatedCost = calculateCost(this.model, usage);
+            if (calculatedCost) {
+                cost = calculatedCost;
+            }
+        }
+        return { usage, cost };
+    }
     buildRequest(config) {
         const messages = config.messages.map((msg) => msg.toOpenAIMessage());
         const request = {
@@ -69,7 +87,9 @@ export class SmolOpenAi extends BaseClient {
                 }
             }
         }
-        return success({ output, toolCalls });
+        // Extract usage and calculate cost
+        const { usage, cost } = this.calculateUsageAndCost(completion.usage);
+        return success({ output, toolCalls, usage, cost });
     }
     async *_textStream(config) {
         const request = this.buildRequest(config);
@@ -77,11 +97,20 @@ export class SmolOpenAi extends BaseClient {
         const completion = await this.client.chat.completions.create({
             ...request,
             stream: true,
+            stream_options: { include_usage: true },
         });
         let content = "";
         const toolCallsMap = new Map();
+        let usage;
+        let cost;
         for await (const chunk of completion) {
             const delta = chunk.choices[0]?.delta;
+            // Extract usage from the final chunk
+            if (chunk.usage) {
+                const usageAndCost = this.calculateUsageAndCost(chunk.usage);
+                usage = usageAndCost.usage;
+                cost = usageAndCost.cost;
+            }
             if (!delta)
                 continue;
             if (delta.content) {
@@ -117,6 +146,9 @@ export class SmolOpenAi extends BaseClient {
             toolCalls.push(toolCall);
             yield { type: "tool_call", toolCall };
         }
-        yield { type: "done", result: { output: content || null, toolCalls } };
+        yield {
+            type: "done",
+            result: { output: content || null, toolCalls, usage, cost },
+        };
     }
 }

package/dist/models.d.ts

@@ -63,11 +63,11 @@ export declare const textModels: readonly [{
 }, {
     readonly type: "text";
     readonly modelName: "o3";
-    readonly description: "o3 is a reasoning model that sets a new standard for math, science, and coding, visual reasoning tasks, and technical writing. Part of the o-series of reasoning models. The knowledge cutoff for o3 models is October, 2023.";
+    readonly description: "o3 is a reasoning model that sets a new standard for math, science, coding, visual reasoning tasks, and technical writing. Part of the o-series of reasoning models. The knowledge cutoff for o3 models is October, 2023.";
     readonly maxInputTokens: 200000;
     readonly maxOutputTokens: 100000;
     readonly inputTokenCost: 2;
-    readonly cachedInputTokenCost: 1;
+    readonly cachedInputTokenCost: 0.5;
     readonly outputTokenCost: 8;
     readonly provider: "openai";
 }, {
@@ -83,7 +83,7 @@ export declare const textModels: readonly [{
 }, {
     readonly type: "text";
     readonly modelName: "o4-mini";
-    readonly description: "o4-mini is a new o-series reasoning model that replaced o3-mini, providing excellent performance for math, science, and coding tasks. Available in ChatGPT Plus, Pro, and Team.";
+    readonly description: "Latest small o-series model optimized for fast, effective reasoning with exceptional performance in coding and visual tasks. Knowledge cutoff: June 2024.";
     readonly maxInputTokens: 200000;
     readonly maxOutputTokens: 100000;
     readonly inputTokenCost: 1.1;
@@ -143,25 +143,25 @@ export declare const textModels: readonly [{
 }, {
     readonly type: "text";
     readonly modelName: "gemini-3-pro-preview";
-    readonly description: "Strongest Gemini 3 model quality with 2M context window. Standard pricing for ≤200k tokens, higher rates for >200k tokens.";
-    readonly maxInputTokens: 2097152;
-    readonly maxOutputTokens: 8192;
+    readonly description: "Strongest Gemini 3 model quality 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). Released Nov 2025, currently in preview.";
+    readonly maxInputTokens: 1048576;
+    readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 2;
     readonly outputTokenCost: 12;
     readonly provider: "google";
 }, {
     readonly type: "text";
     readonly modelName: "gemini-3-flash-preview";
-    readonly description: "Latest Gemini 3 flash model with 1M context window. Excellent performance for high-volume tasks.";
+    readonly description: "Latest Gemini 3 flash model with 1M context window and 64K output. Outperforms 2.5 Pro while being 3x faster. Optimized for agentic workflows and coding. Includes context caching for 90% cost reductions.";
     readonly maxInputTokens: 1048576;
-    readonly maxOutputTokens: 8192;
+    readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 0.5;
     readonly outputTokenCost: 3;
     readonly provider: "google";
 }, {
     readonly type: "text";
     readonly modelName: "gemini-2.5-pro";
-    readonly description: "High-performance Gemini 2.5 model with 2M context window. Standard pricing for ≤200k tokens, higher rates for >200k tokens.";
+    readonly description: "High-performance Gemini 2.5 model with 2M context window. Adaptive thinking for complex reasoning and coding. Standard pricing for ≤200k tokens ($1.25 input/$10.00 output), higher rates for >200k tokens ($2.50 input/higher output). Batch API: 50% discount.";
     readonly maxInputTokens: 2097152;
     readonly maxOutputTokens: 8192;
     readonly inputTokenCost: 1.25;
@@ -170,7 +170,7 @@ export declare const textModels: readonly [{
 }, {
     readonly type: "text";
     readonly modelName: "gemini-2.5-flash";
-    readonly description: "Balanced Gemini 2.5 model with excellent performance-to-cost ratio. 1M context window with free tier available.";
+    readonly description: "Balanced Gemini 2.5 model with excellent performance-to-cost ratio. Lightning-fast with controllable thinking budgets. 1M context window. Context caching available for up to 75% cost reduction.";
     readonly maxInputTokens: 1048576;
     readonly maxOutputTokens: 8192;
     readonly inputTokenCost: 0.3;
@@ -188,11 +188,12 @@ export declare const textModels: readonly [{
 }, {
     readonly type: "text";
     readonly modelName: "gemini-2.0-flash";
-    readonly description: "Workhorse model for all daily tasks. Strong overall performance and supports real-time streaming Live API. 1M context window.";
+    readonly description: "Workhorse model for all daily tasks. Strong overall performance and supports real-time streaming Live API. 1M context window. DEPRECATED: Will be shut down on March 31, 2026.";
     readonly maxInputTokens: 1048576;
     readonly maxOutputTokens: 8192;
     readonly inputTokenCost: 0.1;
     readonly outputTokenCost: 0.4;
+    readonly disabled: true;
     readonly provider: "google";
 }, {
     readonly type: "text";
@@ -316,6 +317,12 @@ export declare const imageModels: readonly [{
     readonly provider: "google";
     readonly description: "aka nano-banana";
     readonly costPerImage: 0.04;
+}, {
+    readonly type: "image";
+    readonly modelName: "gemini-3-pro-image-preview";
+    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;
 }];
 export declare const embeddingsModels: {
     type: string;
@@ -359,11 +366,11 @@ export declare function getModel(modelName: ModelName): {
 } | {
     readonly type: "text";
     readonly modelName: "o3";
-    readonly description: "o3 is a reasoning model that sets a new standard for math, science, and coding, visual reasoning tasks, and technical writing. Part of the o-series of reasoning models. The knowledge cutoff for o3 models is October, 2023.";
+    readonly description: "o3 is a reasoning model that sets a new standard for math, science, coding, visual reasoning tasks, and technical writing. Part of the o-series of reasoning models. The knowledge cutoff for o3 models is October, 2023.";
     readonly maxInputTokens: 200000;
     readonly maxOutputTokens: 100000;
     readonly inputTokenCost: 2;
-    readonly cachedInputTokenCost: 1;
+    readonly cachedInputTokenCost: 0.5;
     readonly outputTokenCost: 8;
     readonly provider: "openai";
 } | {
@@ -379,7 +386,7 @@ export declare function getModel(modelName: ModelName): {
 } | {
     readonly type: "text";
     readonly modelName: "o4-mini";
-    readonly description: "o4-mini is a new o-series reasoning model that replaced o3-mini, providing excellent performance for math, science, and coding tasks. Available in ChatGPT Plus, Pro, and Team.";
+    readonly description: "Latest small o-series model optimized for fast, effective reasoning with exceptional performance in coding and visual tasks. Knowledge cutoff: June 2024.";
     readonly maxInputTokens: 200000;
     readonly maxOutputTokens: 100000;
     readonly inputTokenCost: 1.1;
@@ -439,25 +446,25 @@ export declare function getModel(modelName: ModelName): {
 } | {
     readonly type: "text";
     readonly modelName: "gemini-3-pro-preview";
-    readonly description: "Strongest Gemini 3 model quality with 2M context window. Standard pricing for ≤200k tokens, higher rates for >200k tokens.";
-    readonly maxInputTokens: 2097152;
-    readonly maxOutputTokens: 8192;
+    readonly description: "Strongest Gemini 3 model quality 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). Released Nov 2025, currently in preview.";
+    readonly maxInputTokens: 1048576;
+    readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 2;
     readonly outputTokenCost: 12;
     readonly provider: "google";
 } | {
     readonly type: "text";
     readonly modelName: "gemini-3-flash-preview";
-    readonly description: "Latest Gemini 3 flash model with 1M context window. Excellent performance for high-volume tasks.";
+    readonly description: "Latest Gemini 3 flash model with 1M context window and 64K output. Outperforms 2.5 Pro while being 3x faster. Optimized for agentic workflows and coding. Includes context caching for 90% cost reductions.";
     readonly maxInputTokens: 1048576;
-    readonly maxOutputTokens: 8192;
+    readonly maxOutputTokens: 65536;
     readonly inputTokenCost: 0.5;
     readonly outputTokenCost: 3;
     readonly provider: "google";
 } | {
     readonly type: "text";
     readonly modelName: "gemini-2.5-pro";
-    readonly description: "High-performance Gemini 2.5 model with 2M context window. Standard pricing for ≤200k tokens, higher rates for >200k tokens.";
+    readonly description: "High-performance Gemini 2.5 model with 2M context window. Adaptive thinking for complex reasoning and coding. Standard pricing for ≤200k tokens ($1.25 input/$10.00 output), higher rates for >200k tokens ($2.50 input/higher output). Batch API: 50% discount.";
     readonly maxInputTokens: 2097152;
     readonly maxOutputTokens: 8192;
     readonly inputTokenCost: 1.25;
@@ -466,7 +473,7 @@ export declare function getModel(modelName: ModelName): {
 } | {
     readonly type: "text";
     readonly modelName: "gemini-2.5-flash";
-    readonly description: "Balanced Gemini 2.5 model with excellent performance-to-cost ratio. 1M context window with free tier available.";
+    readonly description: "Balanced Gemini 2.5 model with excellent performance-to-cost ratio. Lightning-fast with controllable thinking budgets. 1M context window. Context caching available for up to 75% cost reduction.";
     readonly maxInputTokens: 1048576;
     readonly maxOutputTokens: 8192;
     readonly inputTokenCost: 0.3;
@@ -484,11 +491,12 @@ export declare function getModel(modelName: ModelName): {
 } | {
     readonly type: "text";
     readonly modelName: "gemini-2.0-flash";
-    readonly description: "Workhorse model for all daily tasks. Strong overall performance and supports real-time streaming Live API. 1M context window.";
+    readonly description: "Workhorse model for all daily tasks. Strong overall performance and supports real-time streaming Live API. 1M context window. DEPRECATED: Will be shut down on March 31, 2026.";
     readonly maxInputTokens: 1048576;
     readonly maxOutputTokens: 8192;
     readonly inputTokenCost: 0.1;
     readonly outputTokenCost: 0.4;
+    readonly disabled: true;
     readonly provider: "google";
 } | {
     readonly type: "text";
@@ -611,8 +619,25 @@ export declare function getModel(modelName: ModelName): {
     readonly provider: "google";
     readonly description: "aka nano-banana";
     readonly costPerImage: 0.04;
+} | {
+    readonly type: "image";
+    readonly modelName: "gemini-3-pro-image-preview";
+    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;
 } | undefined;
 export declare function isImageModel(model: Model): model is ImageModel;
 export declare function isTextModel(model: Model): model is TextModel;
 export declare function isSpeechToTextModel(model: Model): model is SpeechToTextModel;
 export declare function isEmbeddingsModel(model: Model): model is EmbeddingsModel;
+export declare function calculateCost(modelName: ModelName, usage: {
+    inputTokens: number;
+    outputTokens: number;
+    cachedInputTokens?: number;
+}): {
+    inputCost: number;
+    outputCost: number;
+    cachedInputCost?: number;
+    totalCost: number;
+    currency: string;
+} | null;

package/dist/models.js

@@ -1,3 +1,4 @@
+import { round } from "./util.js";
 export const speechToTextModels = [
     { type: "speech-to-text", modelName: "whisper-local", provider: "local" },
     {
@@ -43,11 +44,11 @@ export const textModels = [
     {
         type: "text",
         modelName: "o3",
-        description: "o3 is a reasoning model that sets a new standard for math, science, and coding, visual reasoning tasks, and technical writing. Part of the o-series of reasoning models. The knowledge cutoff for o3 models is October, 2023.",
+        description: "o3 is a reasoning model that sets a new standard for math, science, coding, visual reasoning tasks, and technical writing. Part of the o-series of reasoning models. The knowledge cutoff for o3 models is October, 2023.",
         maxInputTokens: 200000,
         maxOutputTokens: 100000,
         inputTokenCost: 2,
-        cachedInputTokenCost: 1,
+        cachedInputTokenCost: 0.5,
         outputTokenCost: 8,
         provider: "openai",
     },
@@ -65,7 +66,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "o4-mini",
-        description: "o4-mini is a new o-series reasoning model that replaced o3-mini, providing excellent performance for math, science, and coding tasks. Available in ChatGPT Plus, Pro, and Team.",
+        description: "Latest small o-series model optimized for fast, effective reasoning with exceptional performance in coding and visual tasks. Knowledge cutoff: June 2024.",
         maxInputTokens: 200000,
         maxOutputTokens: 100000,
         inputTokenCost: 1.1,
@@ -131,9 +132,9 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-3-pro-preview",
-        description: "Strongest Gemini 3 model quality with 2M context window. Standard pricing for ≤200k tokens, higher rates for >200k tokens.",
-        maxInputTokens: 2_097_152,
-        maxOutputTokens: 8192,
+        description: "Strongest Gemini 3 model quality 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). Released Nov 2025, currently in preview.",
+        maxInputTokens: 1_048_576,
+        maxOutputTokens: 65536,
         inputTokenCost: 2.0,
         outputTokenCost: 12.0,
         provider: "google",
@@ -141,9 +142,9 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-3-flash-preview",
-        description: "Latest Gemini 3 flash model with 1M context window. Excellent performance for high-volume tasks.",
+        description: "Latest Gemini 3 flash model with 1M context window and 64K output. Outperforms 2.5 Pro while being 3x faster. Optimized for agentic workflows and coding. Includes context caching for 90% cost reductions.",
         maxInputTokens: 1_048_576,
-        maxOutputTokens: 8192,
+        maxOutputTokens: 65536,
         inputTokenCost: 0.5,
         outputTokenCost: 3.0,
         provider: "google",
@@ -151,7 +152,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-2.5-pro",
-        description: "High-performance Gemini 2.5 model with 2M context window. Standard pricing for ≤200k tokens, higher rates for >200k tokens.",
+        description: "High-performance Gemini 2.5 model with 2M context window. Adaptive thinking for complex reasoning and coding. Standard pricing for ≤200k tokens ($1.25 input/$10.00 output), higher rates for >200k tokens ($2.50 input/higher output). Batch API: 50% discount.",
         maxInputTokens: 2_097_152,
         maxOutputTokens: 8192,
         inputTokenCost: 1.25,
@@ -161,7 +162,7 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-2.5-flash",
-        description: "Balanced Gemini 2.5 model with excellent performance-to-cost ratio. 1M context window with free tier available.",
+        description: "Balanced Gemini 2.5 model with excellent performance-to-cost ratio. Lightning-fast with controllable thinking budgets. 1M context window. Context caching available for up to 75% cost reduction.",
         maxInputTokens: 1_048_576,
         maxOutputTokens: 8192,
         inputTokenCost: 0.3,
@@ -181,11 +182,12 @@ export const textModels = [
     {
         type: "text",
         modelName: "gemini-2.0-flash",
-        description: "Workhorse model for all daily tasks. Strong overall performance and supports real-time streaming Live API. 1M context window.",
+        description: "Workhorse model for all daily tasks. Strong overall performance and supports real-time streaming Live API. 1M context window. DEPRECATED: Will be shut down on March 31, 2026.",
         maxInputTokens: 1_048_576,
         maxOutputTokens: 8192,
         inputTokenCost: 0.1,
         outputTokenCost: 0.4,
+        disabled: true,
         provider: "google",
     },
     {
@@ -359,6 +361,13 @@ export const imageModels = [
         description: "aka nano-banana",
         costPerImage: 0.04,
     },
+    {
+        type: "image",
+        modelName: "gemini-3-pro-image-preview",
+        provider: "google",
+        description: "High-fidelity image generation with reasoning-enhanced composition. Supports legible text rendering, complex multi-turn editing, and character consistency using up to 14 reference inputs.",
+        costPerImage: 0.05,
+    },
 ];
 export const embeddingsModels = [
     { type: "embeddings", modelName: "text-embedding-3-small", tokenCost: 0.02 },
@@ -379,3 +388,22 @@ export function isSpeechToTextModel(model) {
 export function isEmbeddingsModel(model) {
     return model.type === "embeddings";
 }
+export function calculateCost(modelName, usage) {
+    const model = getModel(modelName);
+    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 cachedInputCost = usage.cachedInputTokens && model.cachedInputTokenCost
+        ? round((usage.cachedInputTokens * model.cachedInputTokenCost) / 1_000_000, 2)
+        : undefined;
+    const totalCost = round(inputCost + outputCost + (cachedInputCost || 0), 2);
+    return {
+        inputCost,
+        outputCost,
+        cachedInputCost,
+        totalCost,
+        currency: "USD",
+    };
+}

package/dist/types.d.ts

@@ -43,9 +43,24 @@ export type ToolLoopDetection = {
     excludeTools?: string[];
 };
 export type BaseClientConfig = SmolConfig & {};
+export type TokenUsage = {
+    inputTokens: number;
+    outputTokens: number;
+    cachedInputTokens?: number;
+    totalTokens?: number;
+};
+export type CostEstimate = {
+    inputCost: number;
+    outputCost: number;
+    cachedInputCost?: number;
+    totalCost: number;
+    currency: string;
+};
 export type PromptResult = {
     output: string | null;
     toolCalls: ToolCall[];
+    usage?: TokenUsage;
+    cost?: CostEstimate;
 };
 export type StreamChunk = {
     type: "text";

package/dist/util.d.ts

@@ -1 +1,2 @@
 export * from "./util/openai.js";
+export declare function round(num: number, places: number): number;

package/dist/util.js

@@ -1 +1,5 @@
 export * from "./util/openai.js";
+export function round(num, places) {
+    const factor = Math.pow(10, places);
+    return Math.round(num * factor) / factor;
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "smoltalk",
-  "version": "0.0.16",
+  "version": "0.0.18",
   "description": "A common interface for LLM APIs",
   "homepage": "https://github.com/egonSchiele/smoltalk",
   "scripts": {