@openrouter/ai-sdk-provider 2.6.0 → 2.7.0

package/dist/internal/index.d.mts

@@ -537,6 +537,40 @@ type OpenRouterSharedSettings = OpenRouterProviderOptions & {
          */
         include: boolean;
     };
+    /**
+     * Default temperature for model calls. Controls randomness in the output.
+     * Can be overridden at call time via generateText/streamText options.
+     * Range: 0 to 2, where 0 is deterministic and higher values are more random.
+     */
+    temperature?: number;
+    /**
+     * Default top-p (nucleus sampling) for model calls.
+     * Can be overridden at call time via generateText/streamText options.
+     * Range: 0 to 1.
+     */
+    topP?: number;
+    /**
+     * Default top-k sampling for model calls.
+     * Can be overridden at call time via generateText/streamText options.
+     */
+    topK?: number;
+    /**
+     * Default frequency penalty for model calls.
+     * Can be overridden at call time via generateText/streamText options.
+     * Range: -2 to 2.
+     */
+    frequencyPenalty?: number;
+    /**
+     * Default presence penalty for model calls.
+     * Can be overridden at call time via generateText/streamText options.
+     * Range: -2 to 2.
+     */
+    presencePenalty?: number;
+    /**
+     * Default maximum number of tokens to generate.
+     * Can be overridden at call time via generateText/streamText options.
+     */
+    maxTokens?: number;
 };
 /**
  * Usage accounting response

package/dist/internal/index.d.ts

@@ -537,6 +537,40 @@ type OpenRouterSharedSettings = OpenRouterProviderOptions & {
          */
         include: boolean;
     };
+    /**
+     * Default temperature for model calls. Controls randomness in the output.
+     * Can be overridden at call time via generateText/streamText options.
+     * Range: 0 to 2, where 0 is deterministic and higher values are more random.
+     */
+    temperature?: number;
+    /**
+     * Default top-p (nucleus sampling) for model calls.
+     * Can be overridden at call time via generateText/streamText options.
+     * Range: 0 to 1.
+     */
+    topP?: number;
+    /**
+     * Default top-k sampling for model calls.
+     * Can be overridden at call time via generateText/streamText options.
+     */
+    topK?: number;
+    /**
+     * Default frequency penalty for model calls.
+     * Can be overridden at call time via generateText/streamText options.
+     * Range: -2 to 2.
+     */
+    frequencyPenalty?: number;
+    /**
+     * Default presence penalty for model calls.
+     * Can be overridden at call time via generateText/streamText options.
+     * Range: -2 to 2.
+     */
+    presencePenalty?: number;
+    /**
+     * Default maximum number of tokens to generate.
+     * Can be overridden at call time via generateText/streamText options.
+     */
+    maxTokens?: number;
 };
 /**
  * Usage accounting response

package/dist/internal/index.js

@@ -3311,7 +3311,7 @@ var OpenRouterChatLanguageModel = class {
     tools,
     toolChoice
   }) {
-    var _a16;
+    var _a16, _b16;
     const baseArgs = __spreadValues(__spreadValues({
       // model id:
       model: this.modelId,
@@ -3322,12 +3322,12 @@ var OpenRouterChatLanguageModel = class {
       top_logprobs: typeof this.settings.logprobs === "number" ? this.settings.logprobs : typeof this.settings.logprobs === "boolean" ? this.settings.logprobs ? 0 : void 0 : void 0,
       user: this.settings.user,
       parallel_tool_calls: this.settings.parallelToolCalls,
-      // standardized settings:
-      max_tokens: maxOutputTokens,
-      temperature,
-      top_p: topP,
-      frequency_penalty: frequencyPenalty,
-      presence_penalty: presencePenalty,
+      // standardized settings (call-level options override model-level settings):
+      max_tokens: maxOutputTokens != null ? maxOutputTokens : this.settings.maxTokens,
+      temperature: temperature != null ? temperature : this.settings.temperature,
+      top_p: topP != null ? topP : this.settings.topP,
+      frequency_penalty: frequencyPenalty != null ? frequencyPenalty : this.settings.frequencyPenalty,
+      presence_penalty: presencePenalty != null ? presencePenalty : this.settings.presencePenalty,
       seed,
       stop: stopSequences,
       response_format: (responseFormat == null ? void 0 : responseFormat.type) === "json" ? responseFormat.schema != null ? {
@@ -3340,7 +3340,7 @@ var OpenRouterChatLanguageModel = class {
           description: responseFormat.description
         })
       } : { type: "json_object" } : void 0,
-      top_k: topK,
+      top_k: topK != null ? topK : this.settings.topK,
       // messages:
       messages: convertToOpenRouterChatMessages(prompt),
       // OpenRouter specific settings:
@@ -3361,14 +3361,18 @@ var OpenRouterChatLanguageModel = class {
       const mappedTools = [];
       for (const tool of tools) {
         if (tool.type === "function") {
-          mappedTools.push({
+          const openrouterOptions = (_b16 = tool.providerOptions) == null ? void 0 : _b16.openrouter;
+          const eagerInputStreaming = openrouterOptions == null ? void 0 : openrouterOptions.eager_input_streaming;
+          mappedTools.push(__spreadValues({
             type: "function",
             function: {
               name: tool.name,
               description: tool.description,
               parameters: tool.inputSchema
             }
-          });
+          }, eagerInputStreaming != null && {
+            eager_input_streaming: eagerInputStreaming
+          }));
         } else if (tool.type === "provider") {
           mappedTools.push(mapProviderTool(tool));
         }
@@ -3706,18 +3710,16 @@ var OpenRouterChatLanguageModel = class {
               return;
             }
             const delta = choice.delta;
-            const emitReasoningChunk = (chunkText, providerMetadata) => {
+            const emitReasoningChunk = (chunkText) => {
               if (!reasoningStarted) {
                 reasoningId = generateId();
                 controller.enqueue({
-                  providerMetadata,
                   type: "reasoning-start",
                   id: reasoningId
                 });
                 reasoningStarted = true;
               }
               controller.enqueue({
-                providerMetadata,
                 type: "reasoning-delta",
                 delta: chunkText,
                 id: reasoningId || generateId()
@@ -3739,15 +3741,10 @@ var OpenRouterChatLanguageModel = class {
                 }
               }
               if (!textStarted) {
-                const reasoningMetadata = {
-                  openrouter: {
-                    reasoning_details: accumulatedReasoningDetails.map((d) => __spreadValues({}, d))
-                  }
-                };
                 for (const detail of delta.reasoning_details) {
                   switch (detail.type) {
                     case "reasoning.text" /* Text */: {
-                      emitReasoningChunk(detail.text || "", reasoningMetadata);
+                      emitReasoningChunk(detail.text || "");
                       break;
                     }
                     case "reasoning.encrypted" /* Encrypted */: {
@@ -3755,7 +3752,7 @@ var OpenRouterChatLanguageModel = class {
                     }
                     case "reasoning.summary" /* Summary */: {
                       if (detail.summary) {
-                        emitReasoningChunk(detail.summary, reasoningMetadata);
+                        emitReasoningChunk(detail.summary);
                       }
                       break;
                     }
@@ -4044,6 +4041,12 @@ var OpenRouterChatLanguageModel = class {
             if (accumulatedFileAnnotations.length > 0) {
               openrouterMetadata.annotations = accumulatedFileAnnotations;
             }
+            if (usage.inputTokens.total === void 0 && openrouterUsage.promptTokens !== void 0) {
+              usage.inputTokens.total = openrouterUsage.promptTokens;
+            }
+            if (usage.outputTokens.total === void 0 && openrouterUsage.completionTokens !== void 0) {
+              usage.outputTokens.total = openrouterUsage.completionTokens;
+            }
             usage.raw = rawUsage;
             controller.enqueue({
               type: "finish",
@@ -4279,16 +4282,16 @@ var OpenRouterCompletionLanguageModel = class {
       logprobs: typeof this.settings.logprobs === "number" ? this.settings.logprobs : typeof this.settings.logprobs === "boolean" ? this.settings.logprobs ? 0 : void 0 : void 0,
       suffix: this.settings.suffix,
       user: this.settings.user,
-      // standardized settings:
-      max_tokens: maxOutputTokens,
-      temperature,
-      top_p: topP,
-      frequency_penalty: frequencyPenalty,
-      presence_penalty: presencePenalty,
+      // standardized settings (call-level options override model-level settings):
+      max_tokens: maxOutputTokens != null ? maxOutputTokens : this.settings.maxTokens,
+      temperature: temperature != null ? temperature : this.settings.temperature,
+      top_p: topP != null ? topP : this.settings.topP,
+      frequency_penalty: frequencyPenalty != null ? frequencyPenalty : this.settings.frequencyPenalty,
+      presence_penalty: presencePenalty != null ? presencePenalty : this.settings.presencePenalty,
       seed,
       stop: stopSequences,
       response_format: responseFormat,
-      top_k: topK,
+      top_k: topK != null ? topK : this.settings.topK,
       // prompt:
       prompt: completionPrompt,
       // OpenRouter specific settings: