@ljoukov/llm 7.0.3 → 7.0.5

package/dist/index.d.cts

@@ -282,12 +282,6 @@ type LlmBaseRequest = {
     readonly imageAspectRatio?: string;
     readonly imageSize?: LlmImageSize;
     readonly thinkingLevel?: LlmThinkingLevel;
-    /**
-     * Gemini-only explicit thinking budget. When provided, this overrides
-     * `thinkingLevel` and the model default for Gemini requests. Use `0` to
-     * disable thinking where the selected Gemini model supports it.
-     */
-    readonly thinkingBudget?: number;
     readonly mediaResolution?: LlmMediaResolution;
     readonly openAiTextFormat?: ResponseTextConfig["format"];
     readonly telemetry?: TelemetrySelection;
@@ -435,7 +429,6 @@ type LlmToolLoopRequest = LlmInput & {
     readonly modelTools?: readonly LlmToolConfig[];
     readonly maxSteps?: number;
     readonly thinkingLevel?: LlmThinkingLevel;
-    readonly thinkingBudget?: number;
     readonly mediaResolution?: LlmMediaResolution;
     readonly steering?: LlmToolLoopSteeringChannel;
     readonly onEvent?: (event: LlmStreamEvent) => void;

package/dist/index.d.ts

@@ -282,12 +282,6 @@ type LlmBaseRequest = {
     readonly imageAspectRatio?: string;
     readonly imageSize?: LlmImageSize;
     readonly thinkingLevel?: LlmThinkingLevel;
-    /**
-     * Gemini-only explicit thinking budget. When provided, this overrides
-     * `thinkingLevel` and the model default for Gemini requests. Use `0` to
-     * disable thinking where the selected Gemini model supports it.
-     */
-    readonly thinkingBudget?: number;
     readonly mediaResolution?: LlmMediaResolution;
     readonly openAiTextFormat?: ResponseTextConfig["format"];
     readonly telemetry?: TelemetrySelection;
@@ -435,7 +429,6 @@ type LlmToolLoopRequest = LlmInput & {
     readonly modelTools?: readonly LlmToolConfig[];
     readonly maxSteps?: number;
     readonly thinkingLevel?: LlmThinkingLevel;
-    readonly thinkingBudget?: number;
     readonly mediaResolution?: LlmMediaResolution;
     readonly steering?: LlmToolLoopSteeringChannel;
     readonly onEvent?: (event: LlmStreamEvent) => void;

package/dist/index.js

@@ -7135,29 +7135,16 @@ function toGeminiThinkingLevel(thinkingLevel) {
       return ThinkingLevel.HIGH;
   }
 }
-function toGemini25ProThinkingBudget(thinkingLevel) {
-  switch (thinkingLevel) {
-    case "low":
-      return 256;
-    case "medium":
-      return 4096;
-    case "high":
-      return 32768;
-  }
-}
-function resolveGeminiThinkingConfig(modelId, thinkingLevel, thinkingBudget) {
+function resolveGeminiThinkingConfig(modelId, thinkingLevel) {
   if (isGeminiImageModelId(modelId) || modelId === "gemini-flash-lite-latest") {
     return void 0;
   }
-  if (thinkingBudget !== void 0) {
-    const normalizedBudget = Math.max(0, Math.floor(thinkingBudget));
-    return normalizedBudget === 0 ? { thinkingBudget: 0 } : { includeThoughts: true, thinkingBudget: normalizedBudget };
-  }
   if (thinkingLevel) {
-    if (modelId === "gemini-2.5-pro") {
+    const thinkingBudget = resolveGeminiThinkingBudget(modelId, thinkingLevel);
+    if (thinkingBudget !== void 0) {
       return {
         includeThoughts: true,
-        thinkingBudget: toGemini25ProThinkingBudget(thinkingLevel)
+        thinkingBudget
       };
     }
     return {
@@ -7179,6 +7166,39 @@ function resolveGeminiThinkingConfig(modelId, thinkingLevel, thinkingBudget) {
       return { includeThoughts: true };
   }
 }
+function resolveGeminiThinkingBudget(modelId, thinkingLevel) {
+  if (modelId === "gemini-2.5-pro") {
+    switch (thinkingLevel) {
+      case "low":
+        return 256;
+      case "medium":
+        return 4096;
+      case "high":
+        return 32768;
+    }
+  }
+  if (modelId === "gemini-flash-latest") {
+    switch (thinkingLevel) {
+      case "low":
+        return 256;
+      case "medium":
+        return 8192;
+      case "high":
+        return 24576;
+    }
+  }
+  if (modelId === "gemini-3-flash-preview") {
+    switch (thinkingLevel) {
+      case "low":
+        return 256;
+      case "medium":
+        return 8192;
+      case "high":
+        return 16384;
+    }
+  }
+  return void 0;
+}
 function decodeInlineDataBuffer(base64) {
   try {
     return Buffer5.from(base64, "base64");
@@ -7528,7 +7548,6 @@ function startLlmCallLoggerFromContents(options) {
       ...options.request.imageAspectRatio ? { imageAspectRatio: options.request.imageAspectRatio } : {},
       ...options.request.imageSize ? { imageSize: options.request.imageSize } : {},
       ...options.request.thinkingLevel ? { thinkingLevel: options.request.thinkingLevel } : {},
-      ...options.request.thinkingBudget !== void 0 ? { thinkingBudget: options.request.thinkingBudget } : {},
       ...options.request.mediaResolution ? { mediaResolution: options.request.mediaResolution } : {},
       ...options.request.openAiTextFormat ? { openAiTextFormat: sanitiseLogValue(options.request.openAiTextFormat) } : {},
       ...getCurrentToolCallContext() ? { toolContext: getCurrentToolCallContext() } : {}
@@ -7830,11 +7849,7 @@ async function runTextCall(params) {
             })
           )
         );
-        const thinkingConfig = resolveGeminiThinkingConfig(
-          modelForProvider,
-          request.thinkingLevel,
-          request.thinkingBudget
-        );
+        const thinkingConfig = resolveGeminiThinkingConfig(modelForProvider, request.thinkingLevel);
         const mediaResolution = toGeminiMediaResolution(request.mediaResolution);
         const config = {
           maxOutputTokens: 32e3,
@@ -8141,7 +8156,6 @@ function startJsonStream(request, operation) {
                 responseMimeType: request.responseMimeType ?? "application/json",
                 responseJsonSchema,
                 thinkingLevel: request.thinkingLevel,
-                thinkingBudget: request.thinkingBudget,
                 ...openAiTextFormatForAttempt ? { openAiTextFormat: openAiTextFormatForAttempt } : {},
                 telemetry: false,
                 signal
@@ -9546,11 +9560,7 @@ async function runToolLoop(request) {
           firstModelEventAtMs = Date.now();
         }
       };
-      const thinkingConfig = resolveGeminiThinkingConfig(
-        request.model,
-        request.thinkingLevel,
-        request.thinkingBudget
-      );
+      const thinkingConfig = resolveGeminiThinkingConfig(request.model, request.thinkingLevel);
       const mediaResolution = toGeminiMediaResolution(request.mediaResolution);
       const config = {
         maxOutputTokens: 32e3,