smoltalk 0.4.0 → 0.4.1

package/dist/clients/anthropic.d.ts

@@ -12,6 +12,13 @@ export declare class SmolAnthropic extends BaseClient implements SmolClient {
     getModel(): ModelName;
     private calculateUsageAndCost;
     private buildRequest;
+    /**
+     * Translate `thinking` / `reasoningEffort` into the correct Anthropic
+     * parameters for this model. Callers shouldn't have to know whether a model
+     * speaks the legacy `budget_tokens` API or the newer adaptive + effort API —
+     * that's exactly the vendor difference this package smooths over.
+     */
+    private resolveThinking;
     private rethrowAsSmolError;
     _textSync(config: SmolConfig): Promise<Result<PromptResult>>;
     _textStream(config: SmolConfig): AsyncGenerator<StreamChunk>;

package/dist/clients/anthropic.js

@@ -6,8 +6,22 @@ import { success, } from "../types.js";
 import { zodToAnthropicTool } from "../util/tool.js";
 import { SmolContentPolicyError, SmolContextWindowExceededError, } from "../smolError.js";
 import { BaseClient } from "./baseClient.js";
+import { getModel, isTextModel } from "../models.js";
 import { Model } from "../model.js";
 const DEFAULT_MAX_TOKENS = 4096;
+/**
+ * Which thinking API a model speaks. New flagship Anthropic models (Opus 4.7+)
+ * reject the legacy `{type: "enabled", budget_tokens}` form with a 400, so we
+ * default unknown/unregistered models to "adaptive" (the forward-looking shape)
+ * rather than the form that's being phased out.
+ */
+function thinkingStyleFor(modelName) {
+    const model = getModel(modelName);
+    if (model && isTextModel(model) && model.reasoning?.thinkingStyle) {
+        return model.reasoning.thinkingStyle;
+    }
+    return "adaptive";
+}
 /**
  * Attach ephemeral cache_control breakpoints to (up to) three places:
  *   1. the last tool definition
@@ -152,28 +166,49 @@ export class SmolAnthropic extends BaseClient {
                 description: tool.description,
             }))
             : undefined;
-        const reasoningBudgetMap = {
-            low: 2048,
-            medium: 5000,
-            high: 10000,
-        };
-        const thinking = config.thinking?.enabled
-            ? {
-                type: "enabled",
-                budget_tokens: config.thinking.budgetTokens ?? 5000,
-            }
-            : config.reasoningEffort
-                ? {
-                    type: "enabled",
-                    budget_tokens: reasoningBudgetMap[config.reasoningEffort],
-                }
-                : undefined;
+        // Normalize the user's provider-agnostic thinking/effort config into the
+        // shape this specific model accepts. Both `thinking.enabled` and
+        // `reasoningEffort` are treated as a request to think.
+        const { thinking, outputConfig } = this.resolveThinking(config);
         const cachingEnabled = config.caching?.enabled !== false;
         const baseRequest = { system, messages: anthropicMessages, tools };
         const finalRequest = cachingEnabled
             ? applyCacheBreakpoints(baseRequest)
             : baseRequest;
-        return { ...finalRequest, thinking };
+        return { ...finalRequest, thinking, outputConfig };
+    }
+    /**
+     * Translate `thinking` / `reasoningEffort` into the correct Anthropic
+     * parameters for this model. Callers shouldn't have to know whether a model
+     * speaks the legacy `budget_tokens` API or the newer adaptive + effort API —
+     * that's exactly the vendor difference this package smooths over.
+     */
+    resolveThinking(config) {
+        const wantsThinking = config.thinking?.enabled === true;
+        const effort = config.reasoningEffort;
+        if (!wantsThinking && !effort) {
+            return { thinking: undefined, outputConfig: undefined };
+        }
+        if (thinkingStyleFor(this.getModel()) === "adaptive") {
+            // `output_config.effort` controls depth; the budget is irrelevant here.
+            return {
+                thinking: { type: "adaptive" },
+                outputConfig: effort ? { effort } : undefined,
+            };
+        }
+        // Legacy budget-based thinking (Haiku 4.5 and older models).
+        const reasoningBudgetMap = {
+            low: 2048,
+            medium: 5000,
+            high: 10000,
+        };
+        const budget_tokens = wantsThinking
+            ? (config.thinking?.budgetTokens ?? 5000)
+            : reasoningBudgetMap[effort];
+        return {
+            thinking: { type: "enabled", budget_tokens },
+            outputConfig: undefined,
+        };
     }
     rethrowAsSmolError(error) {
         if (error instanceof Anthropic.APIError) {
@@ -194,7 +229,7 @@ export class SmolAnthropic extends BaseClient {
         throw error;
     }
     async _textSync(config) {
-        const { system, messages, tools, thinking } = this.buildRequest(config);
+        const { system, messages, tools, thinking, outputConfig } = this.buildRequest(config);
         let debugData = {
             model: this.getModel(),
             max_tokens: config.maxTokens ?? DEFAULT_MAX_TOKENS,
@@ -202,6 +237,7 @@ export class SmolAnthropic extends BaseClient {
             system,
             tools,
             thinking,
+            output_config: outputConfig,
         };
         this.logger.debug("Sending request to Anthropic:", debugData);
         this.statelogClient?.promptRequest(debugData);
@@ -215,6 +251,7 @@ export class SmolAnthropic extends BaseClient {
                 ...(system && { system }),
                 ...(tools && { tools }),
                 ...(thinking && { thinking }),
+                ...(outputConfig && { output_config: outputConfig }),
                 ...(config.temperature !== undefined && {
                     temperature: config.temperature,
                 }),
@@ -253,7 +290,7 @@ export class SmolAnthropic extends BaseClient {
         });
     }
     async *_textStream(config) {
-        const { system, messages, tools, thinking } = this.buildRequest(config);
+        const { system, messages, tools, thinking, outputConfig } = this.buildRequest(config);
         const streamDebugData = {
             model: this.model,
             max_tokens: config.maxTokens ?? DEFAULT_MAX_TOKENS,
@@ -261,6 +298,7 @@ export class SmolAnthropic extends BaseClient {
             system,
             tools,
             thinking,
+            output_config: outputConfig,
         };
         this.logger.debug("Sending streaming request to Anthropic:", streamDebugData);
         this.statelogClient?.promptRequest(streamDebugData);
@@ -274,6 +312,7 @@ export class SmolAnthropic extends BaseClient {
                 ...(system && { system }),
                 ...(tools && { tools }),
                 ...(thinking && { thinking }),
+                ...(outputConfig && { output_config: outputConfig }),
                 ...(config.temperature !== undefined && {
                     temperature: config.temperature,
                 }),

package/dist/models.d.ts

@@ -42,6 +42,16 @@ export type TextModel = BaseModel & {
         levels?: readonly string[];
         /** Default reasoning level */
         defaultLevel?: string;
+        /**
+         * Anthropic thinking API shape:
+         *   "adaptive" → `thinking: {type: "adaptive"}` + `output_config: {effort}`
+         *               (Opus 4.6/4.7/4.8, Sonnet 4.6). Required on 4.7+, where the
+         *               legacy `enabled` form returns a 400.
+         *   "budget"   → `thinking: {type: "enabled", budget_tokens: N}`
+         *               (Haiku 4.5 and older models).
+         * Omit for non-Anthropic models.
+         */
+        thinkingStyle?: "adaptive" | "budget";
         /** Whether reasoning/thinking can be fully disabled */
         canDisable?: boolean;
         /** Whether the response includes visible thinking content (thinking blocks/parts) */
@@ -251,7 +261,7 @@ export declare const textModels: readonly [{
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5-mini";
@@ -269,7 +279,7 @@ export declare const textModels: readonly [{
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5-nano";
@@ -287,7 +297,7 @@ export declare const textModels: readonly [{
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5.1";
@@ -304,7 +314,7 @@ export declare const textModels: readonly [{
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5.2";
@@ -322,7 +332,7 @@ export declare const textModels: readonly [{
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5.2-pro";
@@ -336,7 +346,7 @@ export declare const textModels: readonly [{
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5.4";
@@ -353,7 +363,7 @@ export declare const textModels: readonly [{
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5.4-mini";
@@ -371,7 +381,7 @@ export declare const textModels: readonly [{
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5.4-nano";
@@ -389,7 +399,7 @@ export declare const textModels: readonly [{
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5.4-pro";
@@ -405,7 +415,7 @@ export declare const textModels: readonly [{
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5.5";
@@ -423,7 +433,7 @@ export declare const textModels: readonly [{
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 }, {
     readonly type: "text";
     readonly modelName: "gpt-5.5-pro";
@@ -439,7 +449,7 @@ export declare const textModels: readonly [{
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 }, {
     readonly type: "text";
     readonly modelName: "gemini-3.1-pro-preview";
@@ -652,6 +662,7 @@ export declare const textModels: readonly [{
     readonly cachedInputTokenCost: 0.5;
     readonly outputTokenCost: 25;
     readonly reasoning: {
+        readonly thinkingStyle: "adaptive";
         readonly canDisable: false;
         readonly outputsThinking: true;
         readonly outputsSignatures: true;
@@ -669,6 +680,7 @@ export declare const textModels: readonly [{
     readonly outputTokenCost: 25;
     readonly outputTokensPerSecond: 72;
     readonly reasoning: {
+        readonly thinkingStyle: "adaptive";
         readonly canDisable: false;
         readonly outputsThinking: true;
         readonly outputsSignatures: true;
@@ -686,6 +698,7 @@ export declare const textModels: readonly [{
     readonly outputTokenCost: 25;
     readonly outputTokensPerSecond: 53;
     readonly reasoning: {
+        readonly thinkingStyle: "adaptive";
         readonly canDisable: true;
         readonly outputsThinking: true;
         readonly outputsSignatures: true;
@@ -703,6 +716,7 @@ export declare const textModels: readonly [{
     readonly outputTokenCost: 15;
     readonly outputTokensPerSecond: 52;
     readonly reasoning: {
+        readonly thinkingStyle: "adaptive";
         readonly canDisable: true;
         readonly outputsThinking: true;
         readonly outputsSignatures: true;
@@ -720,6 +734,7 @@ export declare const textModels: readonly [{
     readonly outputTokenCost: 5;
     readonly outputTokensPerSecond: 97;
     readonly reasoning: {
+        readonly thinkingStyle: "budget";
         readonly canDisable: true;
         readonly outputsThinking: true;
         readonly outputsSignatures: true;
@@ -735,6 +750,7 @@ export declare const textModels: readonly [{
     readonly outputTokenCost: 15;
     readonly outputTokensPerSecond: 78;
     readonly reasoning: {
+        readonly thinkingStyle: "budget";
         readonly canDisable: true;
         readonly outputsThinking: true;
         readonly outputsSignatures: true;
@@ -997,7 +1013,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 } | {
     readonly type: "text";
     readonly modelName: "gpt-5-mini";
@@ -1015,7 +1031,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 } | {
     readonly type: "text";
     readonly modelName: "gpt-5-nano";
@@ -1033,7 +1049,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 } | {
     readonly type: "text";
     readonly modelName: "gpt-5.1";
@@ -1050,7 +1066,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 } | {
     readonly type: "text";
     readonly modelName: "gpt-5.2";
@@ -1068,7 +1084,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 } | {
     readonly type: "text";
     readonly modelName: "gpt-5.2-pro";
@@ -1082,7 +1098,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 } | {
     readonly type: "text";
     readonly modelName: "gpt-5.4";
@@ -1099,7 +1115,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 } | {
     readonly type: "text";
     readonly modelName: "gpt-5.4-mini";
@@ -1117,7 +1133,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 } | {
     readonly type: "text";
     readonly modelName: "gpt-5.4-nano";
@@ -1135,7 +1151,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 } | {
     readonly type: "text";
     readonly modelName: "gpt-5.4-pro";
@@ -1151,7 +1167,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 } | {
     readonly type: "text";
     readonly modelName: "gpt-5.5";
@@ -1169,7 +1185,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 } | {
     readonly type: "text";
     readonly modelName: "gpt-5.5-pro";
@@ -1185,7 +1201,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
         readonly outputsThinking: false;
         readonly outputsSignatures: false;
     };
-    readonly provider: "openai";
+    readonly provider: "openai-responses";
 } | {
     readonly type: "text";
     readonly modelName: "gemini-3.1-pro-preview";
@@ -1398,6 +1414,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
     readonly cachedInputTokenCost: 0.5;
     readonly outputTokenCost: 25;
     readonly reasoning: {
+        readonly thinkingStyle: "adaptive";
         readonly canDisable: false;
         readonly outputsThinking: true;
         readonly outputsSignatures: true;
@@ -1415,6 +1432,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
     readonly outputTokenCost: 25;
     readonly outputTokensPerSecond: 72;
     readonly reasoning: {
+        readonly thinkingStyle: "adaptive";
         readonly canDisable: false;
         readonly outputsThinking: true;
         readonly outputsSignatures: true;
@@ -1432,6 +1450,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
     readonly outputTokenCost: 25;
     readonly outputTokensPerSecond: 53;
     readonly reasoning: {
+        readonly thinkingStyle: "adaptive";
         readonly canDisable: true;
         readonly outputsThinking: true;
         readonly outputsSignatures: true;
@@ -1449,6 +1468,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
     readonly outputTokenCost: 15;
     readonly outputTokensPerSecond: 52;
     readonly reasoning: {
+        readonly thinkingStyle: "adaptive";
         readonly canDisable: true;
         readonly outputsThinking: true;
         readonly outputsSignatures: true;
@@ -1466,6 +1486,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
     readonly outputTokenCost: 5;
     readonly outputTokensPerSecond: 97;
     readonly reasoning: {
+        readonly thinkingStyle: "budget";
         readonly canDisable: true;
         readonly outputsThinking: true;
         readonly outputsSignatures: true;
@@ -1481,6 +1502,7 @@ export declare function getModel(modelName: ModelName): TextModel | EmbeddingsMo
     readonly outputTokenCost: 15;
     readonly outputTokensPerSecond: 78;
     readonly reasoning: {
+        readonly thinkingStyle: "budget";
         readonly canDisable: true;
         readonly outputsThinking: true;
         readonly outputsSignatures: true;

package/dist/models.js

@@ -229,7 +229,7 @@ export const textModels = [
             outputsThinking: false,
             outputsSignatures: false,
         },
-        provider: "openai",
+        provider: "openai-responses",
     },
     {
         type: "text",
@@ -248,7 +248,7 @@ export const textModels = [
             outputsThinking: false,
             outputsSignatures: false,
         },
-        provider: "openai",
+        provider: "openai-responses",
     },
     {
         type: "text",
@@ -267,7 +267,7 @@ export const textModels = [
             outputsThinking: false,
             outputsSignatures: false,
         },
-        provider: "openai",
+        provider: "openai-responses",
     },
     {
         type: "text",
@@ -285,7 +285,7 @@ export const textModels = [
             outputsThinking: false,
             outputsSignatures: false,
         },
-        provider: "openai",
+        provider: "openai-responses",
     },
     {
         type: "text",
@@ -304,7 +304,7 @@ export const textModels = [
             outputsThinking: false,
             outputsSignatures: false,
         },
-        provider: "openai",
+        provider: "openai-responses",
     },
     {
         type: "text",
@@ -319,7 +319,7 @@ export const textModels = [
             outputsThinking: false,
             outputsSignatures: false,
         },
-        provider: "openai",
+        provider: "openai-responses",
     },
     {
         type: "text",
@@ -337,7 +337,7 @@ export const textModels = [
             outputsThinking: false,
             outputsSignatures: false,
         },
-        provider: "openai",
+        provider: "openai-responses",
     },
     {
         type: "text",
@@ -356,7 +356,7 @@ export const textModels = [
             outputsThinking: false,
             outputsSignatures: false,
         },
-        provider: "openai",
+        provider: "openai-responses",
     },
     {
         type: "text",
@@ -375,7 +375,7 @@ export const textModels = [
             outputsThinking: false,
             outputsSignatures: false,
         },
-        provider: "openai",
+        provider: "openai-responses",
     },
     {
         type: "text",
@@ -392,7 +392,7 @@ export const textModels = [
             outputsThinking: false,
             outputsSignatures: false,
         },
-        provider: "openai",
+        provider: "openai-responses",
     },
     {
         type: "text",
@@ -411,7 +411,7 @@ export const textModels = [
             outputsThinking: false,
             outputsSignatures: false,
         },
-        provider: "openai",
+        provider: "openai-responses",
     },
     {
         type: "text",
@@ -428,7 +428,7 @@ export const textModels = [
             outputsThinking: false,
             outputsSignatures: false,
         },
-        provider: "openai",
+        provider: "openai-responses",
     },
     {
         type: "text",
@@ -657,6 +657,7 @@ export const textModels = [
         cachedInputTokenCost: 0.5,
         outputTokenCost: 25,
         reasoning: {
+            thinkingStyle: "adaptive",
             canDisable: false,
             outputsThinking: true,
             outputsSignatures: true,
@@ -675,6 +676,7 @@ export const textModels = [
         outputTokenCost: 25,
         outputTokensPerSecond: 72,
         reasoning: {
+            thinkingStyle: "adaptive",
             canDisable: false,
             outputsThinking: true,
             outputsSignatures: true,
@@ -693,6 +695,7 @@ export const textModels = [
         outputTokenCost: 25,
         outputTokensPerSecond: 53,
         reasoning: {
+            thinkingStyle: "adaptive",
             canDisable: true,
             outputsThinking: true,
             outputsSignatures: true,
@@ -711,6 +714,7 @@ export const textModels = [
         outputTokenCost: 15,
         outputTokensPerSecond: 52,
         reasoning: {
+            thinkingStyle: "adaptive",
             canDisable: true,
             outputsThinking: true,
             outputsSignatures: true,
@@ -729,6 +733,7 @@ export const textModels = [
         outputTokenCost: 5,
         outputTokensPerSecond: 97,
         reasoning: {
+            thinkingStyle: "budget",
             canDisable: true,
             outputsThinking: true,
             outputsSignatures: true,
@@ -745,6 +750,7 @@ export const textModels = [
         outputTokenCost: 15,
         outputTokensPerSecond: 78,
         reasoning: {
+            thinkingStyle: "budget",
             canDisable: true,
             outputsThinking: true,
             outputsSignatures: true,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "smoltalk",
-  "version": "0.4.0",
+  "version": "0.4.1",
   "description": "A common interface for LLM APIs",
   "homepage": "https://github.com/egonSchiele/smoltalk",
   "files": [