@ljoukov/llm 7.0.3 → 7.0.5

package/README.md

@@ -491,15 +491,14 @@ const { value } = await generateJson({
 console.log(value.ok, value.message);
 ```
 
-For Gemini calls that need a specific thinking budget, pass `thinkingBudget`. This overrides the model default and
-`thinkingLevel` for Gemini requests:
+For calls that should use a smaller reasoning budget, pass the cross-model `thinkingLevel` option:
 
 ```ts
 const { value } = await generateJson({
   model: "gemini-flash-latest",
   input: "Return a JSON object with ok=true and message='hello'.",
   schema,
-  thinkingBudget: 0,
+  thinkingLevel: "low",
 });
 ```
 

package/dist/index.cjs

@@ -7248,29 +7248,16 @@ function toGeminiThinkingLevel(thinkingLevel) {
       return import_genai2.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 {
@@ -7292,6 +7279,39 @@ 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 import_node_buffer4.Buffer.from(base64, "base64");
@@ -7641,7 +7661,6 @@ 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() } : {}
@@ -7943,11 +7962,7 @@ 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,
@@ -8254,7 +8269,6 @@ function startJsonStream(request, operation) {
                 responseMimeType: request.responseMimeType ?? "application/json",
                 responseJsonSchema,
                 thinkingLevel: request.thinkingLevel,
-                thinkingBudget: request.thinkingBudget,
                 ...openAiTextFormatForAttempt ? { openAiTextFormat: openAiTextFormatForAttempt } : {},
                 telemetry: false,
                 signal
@@ -9659,11 +9673,7 @@ 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,