@ai-sdk/openai 2.0.0-alpha.10 → 2.0.0-alpha.12

package/dist/internal/index.mjs

@@ -264,7 +264,14 @@ var openaiProviderOptions = z.object({
    *
    * @default true
    */
-  structuredOutputs: z.boolean().optional()
+  structuredOutputs: z.boolean().optional(),
+  /**
+   * Service tier for the request. Set to 'flex' for 50% cheaper processing
+   * at the cost of increased latency. Only available for o3 and o4-mini models.
+   *
+   * @default 'auto'
+   */
+  serviceTier: z.enum(["auto", "flex"]).optional()
 });
 
 // src/openai-error.ts
@@ -437,6 +444,7 @@ var OpenAIChatLanguageModel = class {
       metadata: openaiOptions.metadata,
       prediction: openaiOptions.prediction,
       reasoning_effort: openaiOptions.reasoningEffort,
+      service_tier: openaiOptions.serviceTier,
       // messages:
       messages
     };
@@ -510,6 +518,14 @@ var OpenAIChatLanguageModel = class {
         });
       }
     }
+    if (openaiOptions.serviceTier === "flex" && !supportsFlexProcessing(this.modelId)) {
+      warnings.push({
+        type: "unsupported-setting",
+        setting: "serviceTier",
+        details: "flex processing is only available for o3 and o4-mini models"
+      });
+      baseArgs.service_tier = void 0;
+    }
     const {
       tools: openaiTools,
       toolChoice: openaiToolChoice,
@@ -636,6 +652,9 @@ var OpenAIChatLanguageModel = class {
           },
           transform(chunk, controller) {
             var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v, _w, _x;
+            if (options.includeRawChunks) {
+              controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
+            }
             if (!chunk.success) {
               finishReason = "error";
               controller.enqueue({ type: "error", error: chunk.error });
@@ -881,6 +900,9 @@ var openaiChatChunkSchema = z3.union([
 function isReasoningModel(modelId) {
   return modelId.startsWith("o");
 }
+function supportsFlexProcessing(modelId) {
+  return modelId.startsWith("o3") || modelId.startsWith("o4-mini");
+}
 function getSystemMessageMode(modelId) {
   var _a, _b;
   if (!isReasoningModel(modelId)) {
@@ -1221,6 +1243,9 @@ var OpenAICompletionLanguageModel = class {
             controller.enqueue({ type: "stream-start", warnings });
           },
           transform(chunk, controller) {
+            if (options.includeRawChunks) {
+              controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
+            }
             if (!chunk.success) {
               finishReason = "error";
               controller.enqueue({ type: "error", error: chunk.error });
@@ -2116,6 +2141,7 @@ var OpenAIResponsesLanguageModel = class {
       store: openaiOptions == null ? void 0 : openaiOptions.store,
       user: openaiOptions == null ? void 0 : openaiOptions.user,
       instructions: openaiOptions == null ? void 0 : openaiOptions.instructions,
+      service_tier: openaiOptions == null ? void 0 : openaiOptions.serviceTier,
       // model-specific settings:
       ...modelConfig.isReasoningModel && ((openaiOptions == null ? void 0 : openaiOptions.reasoningEffort) != null || (openaiOptions == null ? void 0 : openaiOptions.reasoningSummary) != null) && {
         reasoning: {
@@ -2149,6 +2175,14 @@ var OpenAIResponsesLanguageModel = class {
         });
       }
     }
+    if ((openaiOptions == null ? void 0 : openaiOptions.serviceTier) === "flex" && !supportsFlexProcessing2(this.modelId)) {
+      warnings.push({
+        type: "unsupported-setting",
+        setting: "serviceTier",
+        details: "flex processing is only available for o3 and o4-mini models"
+      });
+      delete baseArgs.service_tier;
+    }
     const {
       tools: openaiTools,
       toolChoice: openaiToolChoice,
@@ -2336,6 +2370,7 @@ var OpenAIResponsesLanguageModel = class {
     let responseId = null;
     const ongoingToolCalls = {};
     let hasToolCalls = false;
+    let lastReasoningSummaryIndex = null;
     return {
       stream: response.pipeThrough(
         new TransformStream({
@@ -2344,6 +2379,9 @@ var OpenAIResponsesLanguageModel = class {
           },
           transform(chunk, controller) {
             var _a, _b, _c, _d, _e, _f, _g, _h;
+            if (options.includeRawChunks) {
+              controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
+            }
             if (!chunk.success) {
               finishReason = "error";
               controller.enqueue({ type: "error", error: chunk.error });
@@ -2389,10 +2427,16 @@ var OpenAIResponsesLanguageModel = class {
                 text: value.delta
               });
             } else if (isResponseReasoningSummaryTextDeltaChunk(value)) {
+              if (lastReasoningSummaryIndex !== null && value.summary_index !== lastReasoningSummaryIndex) {
+                controller.enqueue({ type: "reasoning-part-finish" });
+              }
+              lastReasoningSummaryIndex = value.summary_index;
               controller.enqueue({
                 type: "reasoning",
                 text: value.delta
               });
+            } else if (isResponseReasoningSummaryPartDoneChunk(value)) {
+              controller.enqueue({ type: "reasoning-part-finish" });
             } else if (isResponseOutputItemDoneChunk(value) && value.item.type === "function_call") {
               ongoingToolCalls[value.output_index] = void 0;
               hasToolCalls = true;
@@ -2521,6 +2565,13 @@ var responseReasoningSummaryTextDeltaSchema = z12.object({
   summary_index: z12.number(),
   delta: z12.string()
 });
+var responseReasoningSummaryPartDoneSchema = z12.object({
+  type: z12.literal("response.reasoning_summary_part.done"),
+  item_id: z12.string(),
+  output_index: z12.number(),
+  summary_index: z12.number(),
+  part: z12.unknown().nullish()
+});
 var openaiResponsesChunkSchema = z12.union([
   textDeltaChunkSchema,
   responseFinishedChunkSchema,
@@ -2530,6 +2581,7 @@ var openaiResponsesChunkSchema = z12.union([
   responseOutputItemAddedSchema,
   responseAnnotationAddedSchema,
   responseReasoningSummaryTextDeltaSchema,
+  responseReasoningSummaryPartDoneSchema,
   z12.object({ type: z12.string() }).passthrough()
   // fallback for unknown chunks
 ]);
@@ -2557,6 +2609,9 @@ function isResponseAnnotationAddedChunk(chunk) {
 function isResponseReasoningSummaryTextDeltaChunk(chunk) {
   return chunk.type === "response.reasoning_summary_text.delta";
 }
+function isResponseReasoningSummaryPartDoneChunk(chunk) {
+  return chunk.type === "response.reasoning_summary_part.done";
+}
 function getResponsesModelConfig(modelId) {
   if (modelId.startsWith("o")) {
     if (modelId.startsWith("o1-mini") || modelId.startsWith("o1-preview")) {
@@ -2578,6 +2633,9 @@ function getResponsesModelConfig(modelId) {
     requiredAutoTruncation: false
   };
 }
+function supportsFlexProcessing2(modelId) {
+  return modelId.startsWith("o3") || modelId.startsWith("o4-mini");
+}
 var openaiResponsesProviderOptionsSchema = z12.object({
   metadata: z12.any().nullish(),
   parallelToolCalls: z12.boolean().nullish(),
@@ -2587,7 +2645,8 @@ var openaiResponsesProviderOptionsSchema = z12.object({
   reasoningEffort: z12.string().nullish(),
   strictSchemas: z12.boolean().nullish(),
   instructions: z12.string().nullish(),
-  reasoningSummary: z12.string().nullish()
+  reasoningSummary: z12.string().nullish(),
+  serviceTier: z12.enum(["auto", "flex"]).nullish()
 });
 export {
   OpenAIChatLanguageModel,