smoltalk 0.3.1 → 0.4.1

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",
@@ -459,10 +459,28 @@ export const textModels = [
         disabled: true,
         provider: "google",
     },
+    {
+        type: "text",
+        modelName: "gemini-3.5-flash",
+        description: "Latest Gemini 3.5 Flash model (GA May 2026). Outperforms Gemini 3.1 Pro on coding and agentic suites at 4x the speed. 1M context window, 64K output. Context caching: $0.15/1M read.",
+        maxInputTokens: 1_048_576,
+        maxOutputTokens: 65536,
+        inputTokenCost: 1.5,
+        cachedInputTokenCost: 0.15,
+        outputTokenCost: 9.0,
+        reasoning: {
+            levels: ["minimal", "low", "medium", "high"],
+            defaultLevel: "high",
+            canDisable: false,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
+        provider: "google",
+    },
     {
         type: "text",
         modelName: "gemini-3-flash-preview",
-        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.",
+        description: "Gemini 3 Flash preview. Superseded by gemini-3.5-flash. 1M context window and 64K output. Optimized for agentic workflows and coding.",
         maxInputTokens: 1_048_576,
         maxOutputTokens: 65536,
         inputTokenCost: 0.5,
@@ -479,8 +497,8 @@ export const textModels = [
     },
     {
         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.",
+        modelName: "gemini-3.1-flash-lite",
+        description: "Most cost-effective Gemini 3.1 model (GA). Thinking support, 1M context window, 64K output. 2.5x faster TTFA and 45% faster output than 2.5 Flash.",
         maxInputTokens: 1_048_576,
         maxOutputTokens: 65536,
         inputTokenCost: 0.25,
@@ -495,6 +513,17 @@ export const textModels = [
         },
         provider: "google",
     },
+    {
+        type: "text",
+        modelName: "gemini-3.1-flash-lite-preview",
+        description: "DEPRECATED: Preview version, discontinued July 9, 2026. Use gemini-3.1-flash-lite instead.",
+        maxInputTokens: 1_048_576,
+        maxOutputTokens: 65536,
+        inputTokenCost: 0.25,
+        outputTokenCost: 1.5,
+        disabled: true,
+        provider: "google",
+    },
     {
         type: "text",
         modelName: "gemini-2.5-pro",
@@ -502,6 +531,7 @@ export const textModels = [
         maxInputTokens: 2_097_152,
         maxOutputTokens: 65536,
         inputTokenCost: 1.25,
+        cachedInputTokenCost: 0.31,
         outputTokenCost: 10.0,
         outputTokensPerSecond: 145,
         reasoning: {
@@ -518,6 +548,7 @@ export const textModels = [
         maxInputTokens: 1_048_576,
         maxOutputTokens: 65536,
         inputTokenCost: 0.3,
+        cachedInputTokenCost: 0.075,
         outputTokenCost: 2.5,
         outputTokensPerSecond: 245,
         reasoning: {
@@ -534,6 +565,7 @@ export const textModels = [
         maxInputTokens: 1_048_576,
         maxOutputTokens: 65536,
         inputTokenCost: 0.1,
+        cachedInputTokenCost: 0.025,
         outputTokenCost: 0.4,
         outputTokensPerSecond: 400,
         reasoning: {
@@ -615,17 +647,36 @@ export const textModels = [
         disabled: true,
         provider: "google",
     },
+    {
+        type: "text",
+        modelName: "claude-opus-4-8",
+        description: "The most capable Claude model for complex reasoning and agentic coding. Same per-token pricing as Opus 4.7 with improved tool-use efficiency (~290 tokens for tool-use system prompt vs 675 on 4.7). 1M context window, 128K max output.",
+        maxInputTokens: 1_000_000,
+        maxOutputTokens: 128_000,
+        inputTokenCost: 5,
+        cachedInputTokenCost: 0.5,
+        outputTokenCost: 25,
+        reasoning: {
+            thinkingStyle: "adaptive",
+            canDisable: false,
+            outputsThinking: true,
+            outputsSignatures: true,
+        },
+        provider: "anthropic",
+    },
     {
         type: "text",
         modelName: "claude-opus-4-7",
-        description: "The most capable Claude model for complex reasoning and agentic coding. Features Adaptive Thinking that auto-tunes reasoning compute per request. 1M context window, 128K max output. Knowledge cutoff: January 2026.",
+        description: "Claude Opus 4.7 for complex reasoning and agentic coding. Features Adaptive Thinking that auto-tunes reasoning compute per request. 1M context window, 128K max output. Knowledge cutoff: January 2026.",
         maxInputTokens: 1_000_000,
         maxOutputTokens: 128_000,
         inputTokenCost: 5,
         cachedInputTokenCost: 0.5,
+        cacheCreationInputTokenCost: 6.25,
         outputTokenCost: 25,
         outputTokensPerSecond: 72,
         reasoning: {
+            thinkingStyle: "adaptive",
             canDisable: false,
             outputsThinking: true,
             outputsSignatures: true,
@@ -640,9 +691,11 @@ export const textModels = [
         maxOutputTokens: 128_000,
         inputTokenCost: 5,
         cachedInputTokenCost: 0.5,
+        cacheCreationInputTokenCost: 6.25,
         outputTokenCost: 25,
         outputTokensPerSecond: 53,
         reasoning: {
+            thinkingStyle: "adaptive",
             canDisable: true,
             outputsThinking: true,
             outputsSignatures: true,
@@ -657,9 +710,11 @@ export const textModels = [
         maxOutputTokens: 64_000,
         inputTokenCost: 3,
         cachedInputTokenCost: 0.3,
+        cacheCreationInputTokenCost: 3.75,
         outputTokenCost: 15,
         outputTokensPerSecond: 52,
         reasoning: {
+            thinkingStyle: "adaptive",
             canDisable: true,
             outputsThinking: true,
             outputsSignatures: true,
@@ -674,9 +729,11 @@ export const textModels = [
         maxOutputTokens: 64_000,
         inputTokenCost: 1,
         cachedInputTokenCost: 0.1,
+        cacheCreationInputTokenCost: 1.25,
         outputTokenCost: 5,
         outputTokensPerSecond: 97,
         reasoning: {
+            thinkingStyle: "budget",
             canDisable: true,
             outputsThinking: true,
             outputsSignatures: true,
@@ -693,6 +750,7 @@ export const textModels = [
         outputTokenCost: 15,
         outputTokensPerSecond: 78,
         reasoning: {
+            thinkingStyle: "budget",
             canDisable: true,
             outputsThinking: true,
             outputsSignatures: true,
@@ -751,7 +809,15 @@ export const imageModels = [
         type: "image",
         modelName: "gemini-3.1-flash-image-preview",
         provider: "google",
-        description: "Fast image generation with Gemini 3.1 Flash. Supports resolutions from 512px to 4096px. ~$0.067/image at 1K resolution.",
+        description: "DEPRECATED: Preview version. Use gemini-3.1-flash-image instead.",
+        costPerImage: 0.067,
+        disabled: true,
+    },
+    {
+        type: "image",
+        modelName: "gemini-3.1-flash-image",
+        provider: "google",
+        description: "Fast image generation with Gemini 3.1 Flash (GA). Supports resolutions from 512px to 4096px. ~$0.045/image at 512px, $0.067 at 1K, $0.101 at 2K, $0.151 at 4K.",
         costPerImage: 0.067,
     },
 ];

package/dist/types/costEstimate.d.ts

@@ -3,6 +3,7 @@ export type CostEstimate = {
     inputCost: number;
     outputCost: number;
     cachedInputCost?: number;
+    cacheCreationInputCost?: number;
     totalCost: number;
     currency: string;
 };
@@ -10,6 +11,7 @@ export declare const CostEstimateSchema: z.ZodObject<{
     inputCost: z.ZodNumber;
     outputCost: z.ZodNumber;
     cachedInputCost: z.ZodOptional<z.ZodNumber>;
+    cacheCreationInputCost: z.ZodOptional<z.ZodNumber>;
     totalCost: z.ZodNumber;
     currency: z.ZodString;
 }, z.core.$strip>;

package/dist/types/costEstimate.js

@@ -3,6 +3,7 @@ export const CostEstimateSchema = z.object({
     inputCost: z.number(),
     outputCost: z.number(),
     cachedInputCost: z.number().optional(),
+    cacheCreationInputCost: z.number().optional(),
     totalCost: z.number(),
     currency: z.string(),
 });
@@ -24,6 +25,7 @@ export function addCosts(_a, _b) {
         inputCost: a.inputCost + b.inputCost,
         outputCost: a.outputCost + b.outputCost,
         cachedInputCost: (a.cachedInputCost || 0) + (b.cachedInputCost || 0),
+        cacheCreationInputCost: (a.cacheCreationInputCost || 0) + (b.cacheCreationInputCost || 0),
         totalCost: a.totalCost + b.totalCost,
         currency: a.currency,
     };

package/dist/types/tokenUsage.d.ts

@@ -3,12 +3,14 @@ export type TokenUsage = {
     inputTokens: number;
     outputTokens: number;
     cachedInputTokens?: number;
+    cacheCreationInputTokens?: number;
     totalTokens?: number;
 };
 export declare const TokenUsageSchema: z.ZodObject<{
     inputTokens: z.ZodNumber;
     outputTokens: z.ZodNumber;
     cachedInputTokens: z.ZodOptional<z.ZodNumber>;
+    cacheCreationInputTokens: z.ZodOptional<z.ZodNumber>;
     totalTokens: z.ZodOptional<z.ZodNumber>;
 }, z.core.$strip>;
 export declare function addTokenUsage(_a?: TokenUsage, _b?: TokenUsage): TokenUsage;

package/dist/types/tokenUsage.js

@@ -3,6 +3,7 @@ export const TokenUsageSchema = z.object({
     inputTokens: z.number(),
     outputTokens: z.number(),
     cachedInputTokens: z.number().optional(),
+    cacheCreationInputTokens: z.number().optional(),
     totalTokens: z.number().optional(),
 });
 export function addTokenUsage(_a, _b) {
@@ -20,6 +21,7 @@ export function addTokenUsage(_a, _b) {
         inputTokens: a.inputTokens + b.inputTokens,
         outputTokens: a.outputTokens + b.outputTokens,
         cachedInputTokens: (a.cachedInputTokens || 0) + (b.cachedInputTokens || 0),
+        cacheCreationInputTokens: (a.cacheCreationInputTokens || 0) + (b.cacheCreationInputTokens || 0),
         totalTokens: (a.totalTokens || 0) + (b.totalTokens || 0),
     };
 }

package/dist/types.d.ts

@@ -69,6 +69,10 @@ export type SmolConfig = {
         enabled: boolean;
         budgetTokens?: number;
     };
+    /** Prompt caching. Currently used by Anthropic; OpenAI/Google cache automatically. Defaults to enabled. */
+    caching?: {
+        enabled?: boolean;
+    };
     /** Provider-agnostic reasoning effort level. */
     reasoningEffort?: "low" | "medium" | "high";
     responseFormatOptions?: Partial<{

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "smoltalk",
-  "version": "0.3.1",
+  "version": "0.4.1",
   "description": "A common interface for LLM APIs",
   "homepage": "https://github.com/egonSchiele/smoltalk",
   "files": [