@ljoukov/llm 3.0.14 → 4.0.0

package/dist/index.cjs

@@ -199,9 +199,14 @@ var FIREWORKS_GLM_5_PRICING = {
 };
 var FIREWORKS_MINIMAX_M21_PRICING = {
   inputRate: 0.3 / 1e6,
-  cachedRate: 0.15 / 1e6,
+  cachedRate: 0.03 / 1e6,
   outputRate: 1.2 / 1e6
 };
+var FIREWORKS_GPT_OSS_120B_PRICING = {
+  inputRate: 0.15 / 1e6,
+  cachedRate: 0.075 / 1e6,
+  outputRate: 0.6 / 1e6
+};
 function getFireworksPricing(modelId) {
   if (modelId.includes("kimi-k2.5") || modelId.includes("kimi-k2p5")) {
     return FIREWORKS_KIMI_K25_PRICING;
@@ -212,6 +217,9 @@ function getFireworksPricing(modelId) {
   if (modelId.includes("minimax-m2.1") || modelId.includes("minimax-m2p1")) {
     return FIREWORKS_MINIMAX_M21_PRICING;
   }
+  if (modelId.includes("gpt-oss-120b")) {
+    return FIREWORKS_GPT_OSS_120B_PRICING;
+  }
   return void 0;
 }
 
@@ -243,7 +251,16 @@ var GEMINI_2_5_FLASH_PRICING = {
   outputRateLow: 2.5 / 1e6,
   outputRateHigh: 2.5 / 1e6
 };
-var GEMINI_IMAGE_PREVIEW_PRICING = {
+var GEMINI_2_5_FLASH_LITE_PRICING = {
+  threshold: 2e5,
+  inputRateLow: 0.1 / 1e6,
+  inputRateHigh: 0.1 / 1e6,
+  cachedRateLow: 0.025 / 1e6,
+  cachedRateHigh: 0.025 / 1e6,
+  outputRateLow: 0.4 / 1e6,
+  outputRateHigh: 0.4 / 1e6
+};
+var GEMINI_3_PRO_IMAGE_PREVIEW_PRICING = {
   inputRate: 2 / 1e6,
   cachedRate: 0.2 / 1e6,
   outputTextRate: 12 / 1e6,
@@ -254,11 +271,26 @@ var GEMINI_IMAGE_PREVIEW_PRICING = {
     "4K": 0.24
   }
 };
+var GEMINI_3_1_FLASH_IMAGE_PREVIEW_PRICING = {
+  inputRate: 0.5 / 1e6,
+  cachedRate: 0.125 / 1e6,
+  outputTextRate: 3 / 1e6,
+  outputImageRate: 60 / 1e6,
+  imagePrices: {
+    "512": 0.045,
+    "1K": 0.067,
+    "2K": 0.101,
+    "4K": 0.15
+  }
+};
 function getGeminiProPricing(modelId) {
   if (modelId.includes("gemini-2.5-pro")) {
     return GEMINI_2_5_PRO_PRICING;
   }
-  if (modelId.includes("gemini-2.5-flash") || modelId.includes("gemini-flash-latest")) {
+  if (modelId.includes("gemini-flash-lite-latest")) {
+    return GEMINI_2_5_FLASH_LITE_PRICING;
+  }
+  if (modelId.includes("gemini-2.5-flash") || modelId.includes("gemini-flash-latest") || modelId.includes("gemini-3-flash-preview")) {
     return GEMINI_2_5_FLASH_PRICING;
   }
   if (modelId.includes("gemini-3-pro") || modelId.includes("gemini-3.1-pro")) {
@@ -267,8 +299,14 @@ function getGeminiProPricing(modelId) {
   return void 0;
 }
 function getGeminiImagePricing(modelId) {
+  if (modelId.includes("gemini-3.1-flash-image-preview")) {
+    return GEMINI_3_1_FLASH_IMAGE_PREVIEW_PRICING;
+  }
+  if (modelId.includes("gemini-3-pro-image-preview")) {
+    return GEMINI_3_PRO_IMAGE_PREVIEW_PRICING;
+  }
   if (modelId.includes("image-preview")) {
-    return GEMINI_IMAGE_PREVIEW_PRICING;
+    return GEMINI_3_PRO_IMAGE_PREVIEW_PRICING;
   }
   return void 0;
 }
@@ -3116,9 +3154,16 @@ function resolveProvider(model) {
 function isOpenAiCodexModel(modelId) {
   return modelId.includes("codex");
 }
-function resolveOpenAiReasoningEffort(modelId, override) {
-  if (override) {
-    return override;
+function resolveOpenAiReasoningEffort(modelId, thinkingLevel) {
+  if (thinkingLevel) {
+    switch (thinkingLevel) {
+      case "low":
+        return "low";
+      case "medium":
+        return "medium";
+      case "high":
+        return "xhigh";
+    }
   }
   if (isOpenAiCodexModel(modelId)) {
     return "medium";
@@ -4403,10 +4448,42 @@ function extractFireworksToolCalls(message) {
   }
   return calls;
 }
-function resolveGeminiThinkingConfig(modelId) {
+function toGeminiThinkingLevel(thinkingLevel) {
+  switch (thinkingLevel) {
+    case "low":
+      return import_genai2.ThinkingLevel.LOW;
+    case "medium":
+      return import_genai2.ThinkingLevel.MEDIUM;
+    case "high":
+      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) {
   if (isGeminiImageModelId(modelId)) {
     return void 0;
   }
+  if (thinkingLevel) {
+    if (modelId === "gemini-2.5-pro") {
+      return {
+        includeThoughts: true,
+        thinkingBudget: toGemini25ProThinkingBudget(thinkingLevel)
+      };
+    }
+    return {
+      includeThoughts: true,
+      thinkingLevel: toGeminiThinkingLevel(thinkingLevel)
+    };
+  }
   switch (modelId) {
     case "gemini-3.1-pro-preview":
       return { includeThoughts: true };
@@ -4493,10 +4570,7 @@ async function runTextCall(params) {
   if (provider === "openai") {
     const openAiInput = toOpenAiInput(contents);
     const openAiTools = toOpenAiTools(request.tools);
-    const reasoningEffort = resolveOpenAiReasoningEffort(
-      modelForProvider,
-      request.openAiReasoningEffort
-    );
+    const reasoningEffort = resolveOpenAiReasoningEffort(modelForProvider, request.thinkingLevel);
     const openAiTextConfig = {
       format: request.openAiTextFormat ?? { type: "text" },
       verbosity: resolveOpenAiVerbosity(modelForProvider)
@@ -4566,10 +4640,7 @@ async function runTextCall(params) {
     }, modelForProvider);
   } else if (provider === "chatgpt") {
     const chatGptInput = toChatGptInput(contents);
-    const reasoningEffort = resolveOpenAiReasoningEffort(
-      request.model,
-      request.openAiReasoningEffort
-    );
+    const reasoningEffort = resolveOpenAiReasoningEffort(request.model, request.thinkingLevel);
     const openAiTools = toOpenAiTools(request.tools);
     const requestPayload = {
       model: modelForProvider,
@@ -4661,7 +4732,7 @@ async function runTextCall(params) {
     }, modelForProvider);
   } else {
     const geminiContents = contents.map(convertLlmContentToGeminiContent);
-    const thinkingConfig = resolveGeminiThinkingConfig(modelForProvider);
+    const thinkingConfig = resolveGeminiThinkingConfig(modelForProvider, request.thinkingLevel);
     const config = {
       maxOutputTokens: 32e3,
       ...thinkingConfig ? { thinkingConfig } : {},
@@ -4839,7 +4910,7 @@ function streamJson(request) {
           tools: request.tools,
           responseMimeType: request.responseMimeType ?? "application/json",
           responseJsonSchema,
-          openAiReasoningEffort: request.openAiReasoningEffort,
+          thinkingLevel: request.thinkingLevel,
           ...openAiTextFormatForAttempt ? { openAiTextFormat: openAiTextFormatForAttempt } : {},
           signal
         });
@@ -4916,7 +4987,7 @@ async function generateJson(request) {
         tools: request.tools,
         responseMimeType: request.responseMimeType ?? "application/json",
         responseJsonSchema,
-        openAiReasoningEffort: request.openAiReasoningEffort,
+        thinkingLevel: request.thinkingLevel,
         ...openAiTextFormatForAttempt ? { openAiTextFormat: openAiTextFormatForAttempt } : {},
         signal: request.signal
       });
@@ -5182,7 +5253,7 @@ async function runToolLoop(request) {
       const openAiTools = openAiNativeTools ? [...openAiNativeTools, ...openAiAgentTools] : [...openAiAgentTools];
       const reasoningEffort = resolveOpenAiReasoningEffort(
         providerInfo.model,
-        request.openAiReasoningEffort
+        request.thinkingLevel
       );
       const textConfig = {
         format: { type: "text" },
@@ -5450,10 +5521,7 @@ async function runToolLoop(request) {
       const openAiAgentTools = buildOpenAiToolsFromToolSet(request.tools);
       const openAiNativeTools = toOpenAiTools(request.modelTools);
       const openAiTools = openAiNativeTools ? [...openAiNativeTools, ...openAiAgentTools] : [...openAiAgentTools];
-      const reasoningEffort = resolveOpenAiReasoningEffort(
-        request.model,
-        request.openAiReasoningEffort
-      );
+      const reasoningEffort = resolveOpenAiReasoningEffort(request.model, request.thinkingLevel);
       const toolLoopInput = toChatGptInput(contents);
       const conversationId = `tool-loop-${(0, import_node_crypto.randomBytes)(8).toString("hex")}`;
       const promptCacheKey = conversationId;
@@ -5892,7 +5960,7 @@ async function runToolLoop(request) {
           firstModelEventAtMs = Date.now();
         }
       };
-      const thinkingConfig = resolveGeminiThinkingConfig(request.model);
+      const thinkingConfig = resolveGeminiThinkingConfig(request.model, request.thinkingLevel);
       const config = {
         maxOutputTokens: 32e3,
         tools: geminiTools,
@@ -9268,7 +9336,7 @@ function createSubagentController(params) {
           subagentTool: params.subagentSelection,
           modelTools: params.toolLoopRequest.modelTools,
           maxSteps: subagentRequest.maxSteps,
-          openAiReasoningEffort: params.toolLoopRequest.openAiReasoningEffort,
+          thinkingLevel: params.toolLoopRequest.thinkingLevel,
           signal: subagentRequest.signal
         },
         {