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

package/dist/internal/index.d.mts

@@ -56,6 +56,13 @@ declare const openaiProviderOptions: z.ZodObject<{
      * @default true
      */
     structuredOutputs: z.ZodOptional<z.ZodBoolean>;
+    /**
+     * 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.ZodOptional<z.ZodEnum<["auto", "flex"]>>;
 }, "strip", z.ZodTypeAny, {
     user?: string | undefined;
     logitBias?: Record<number, number> | undefined;
@@ -67,6 +74,7 @@ declare const openaiProviderOptions: z.ZodObject<{
     metadata?: Record<string, string> | undefined;
     prediction?: Record<string, any> | undefined;
     structuredOutputs?: boolean | undefined;
+    serviceTier?: "auto" | "flex" | undefined;
 }, {
     user?: string | undefined;
     logitBias?: Record<number, number> | undefined;
@@ -78,6 +86,7 @@ declare const openaiProviderOptions: z.ZodObject<{
     metadata?: Record<string, string> | undefined;
     prediction?: Record<string, any> | undefined;
     structuredOutputs?: boolean | undefined;
+    serviceTier?: "auto" | "flex" | undefined;
 }>;
 type OpenAIProviderOptions = z.infer<typeof openaiProviderOptions>;
 
@@ -353,12 +362,14 @@ declare const openaiResponsesProviderOptionsSchema: z.ZodObject<{
     strictSchemas: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
     instructions: z.ZodOptional<z.ZodNullable<z.ZodString>>;
     reasoningSummary: z.ZodOptional<z.ZodNullable<z.ZodString>>;
+    serviceTier: z.ZodOptional<z.ZodNullable<z.ZodEnum<["auto", "flex"]>>>;
 }, "strip", z.ZodTypeAny, {
     user?: string | null | undefined;
     parallelToolCalls?: boolean | null | undefined;
     reasoningEffort?: string | null | undefined;
     store?: boolean | null | undefined;
     metadata?: any;
+    serviceTier?: "auto" | "flex" | null | undefined;
     instructions?: string | null | undefined;
     previousResponseId?: string | null | undefined;
     strictSchemas?: boolean | null | undefined;
@@ -369,6 +380,7 @@ declare const openaiResponsesProviderOptionsSchema: z.ZodObject<{
     reasoningEffort?: string | null | undefined;
     store?: boolean | null | undefined;
     metadata?: any;
+    serviceTier?: "auto" | "flex" | null | undefined;
     instructions?: string | null | undefined;
     previousResponseId?: string | null | undefined;
     strictSchemas?: boolean | null | undefined;

package/dist/internal/index.d.ts

@@ -56,6 +56,13 @@ declare const openaiProviderOptions: z.ZodObject<{
      * @default true
      */
     structuredOutputs: z.ZodOptional<z.ZodBoolean>;
+    /**
+     * 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.ZodOptional<z.ZodEnum<["auto", "flex"]>>;
 }, "strip", z.ZodTypeAny, {
     user?: string | undefined;
     logitBias?: Record<number, number> | undefined;
@@ -67,6 +74,7 @@ declare const openaiProviderOptions: z.ZodObject<{
     metadata?: Record<string, string> | undefined;
     prediction?: Record<string, any> | undefined;
     structuredOutputs?: boolean | undefined;
+    serviceTier?: "auto" | "flex" | undefined;
 }, {
     user?: string | undefined;
     logitBias?: Record<number, number> | undefined;
@@ -78,6 +86,7 @@ declare const openaiProviderOptions: z.ZodObject<{
     metadata?: Record<string, string> | undefined;
     prediction?: Record<string, any> | undefined;
     structuredOutputs?: boolean | undefined;
+    serviceTier?: "auto" | "flex" | undefined;
 }>;
 type OpenAIProviderOptions = z.infer<typeof openaiProviderOptions>;
 
@@ -353,12 +362,14 @@ declare const openaiResponsesProviderOptionsSchema: z.ZodObject<{
     strictSchemas: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
     instructions: z.ZodOptional<z.ZodNullable<z.ZodString>>;
     reasoningSummary: z.ZodOptional<z.ZodNullable<z.ZodString>>;
+    serviceTier: z.ZodOptional<z.ZodNullable<z.ZodEnum<["auto", "flex"]>>>;
 }, "strip", z.ZodTypeAny, {
     user?: string | null | undefined;
     parallelToolCalls?: boolean | null | undefined;
     reasoningEffort?: string | null | undefined;
     store?: boolean | null | undefined;
     metadata?: any;
+    serviceTier?: "auto" | "flex" | null | undefined;
     instructions?: string | null | undefined;
     previousResponseId?: string | null | undefined;
     strictSchemas?: boolean | null | undefined;
@@ -369,6 +380,7 @@ declare const openaiResponsesProviderOptionsSchema: z.ZodObject<{
     reasoningEffort?: string | null | undefined;
     store?: boolean | null | undefined;
     metadata?: any;
+    serviceTier?: "auto" | "flex" | null | undefined;
     instructions?: string | null | undefined;
     previousResponseId?: string | null | undefined;
     strictSchemas?: boolean | null | undefined;

package/dist/internal/index.js

@@ -290,7 +290,14 @@ var openaiProviderOptions = import_zod.z.object({
    *
    * @default true
    */
-  structuredOutputs: import_zod.z.boolean().optional()
+  structuredOutputs: import_zod.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: import_zod.z.enum(["auto", "flex"]).optional()
 });
 
 // src/openai-error.ts
@@ -461,6 +468,7 @@ var OpenAIChatLanguageModel = class {
       metadata: openaiOptions.metadata,
       prediction: openaiOptions.prediction,
       reasoning_effort: openaiOptions.reasoningEffort,
+      service_tier: openaiOptions.serviceTier,
       // messages:
       messages
     };
@@ -534,6 +542,14 @@ 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,
@@ -660,6 +676,9 @@ 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 });
@@ -905,6 +924,9 @@ var openaiChatChunkSchema = import_zod3.z.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)) {
@@ -1236,6 +1258,9 @@ 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 });
@@ -2098,6 +2123,7 @@ 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: {
@@ -2131,6 +2157,14 @@ 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,
@@ -2318,6 +2352,7 @@ var OpenAIResponsesLanguageModel = class {
     let responseId = null;
     const ongoingToolCalls = {};
     let hasToolCalls = false;
+    let lastReasoningSummaryIndex = null;
     return {
       stream: response.pipeThrough(
         new TransformStream({
@@ -2326,6 +2361,9 @@ 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 });
@@ -2371,10 +2409,16 @@ 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;
@@ -2503,6 +2547,13 @@ var responseReasoningSummaryTextDeltaSchema = import_zod12.z.object({
   summary_index: import_zod12.z.number(),
   delta: import_zod12.z.string()
 });
+var responseReasoningSummaryPartDoneSchema = import_zod12.z.object({
+  type: import_zod12.z.literal("response.reasoning_summary_part.done"),
+  item_id: import_zod12.z.string(),
+  output_index: import_zod12.z.number(),
+  summary_index: import_zod12.z.number(),
+  part: import_zod12.z.unknown().nullish()
+});
 var openaiResponsesChunkSchema = import_zod12.z.union([
   textDeltaChunkSchema,
   responseFinishedChunkSchema,
@@ -2512,6 +2563,7 @@ var openaiResponsesChunkSchema = import_zod12.z.union([
   responseOutputItemAddedSchema,
   responseAnnotationAddedSchema,
   responseReasoningSummaryTextDeltaSchema,
+  responseReasoningSummaryPartDoneSchema,
   import_zod12.z.object({ type: import_zod12.z.string() }).passthrough()
   // fallback for unknown chunks
 ]);
@@ -2539,6 +2591,9 @@ 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")) {
@@ -2560,6 +2615,9 @@ function getResponsesModelConfig(modelId) {
     requiredAutoTruncation: false
   };
 }
+function supportsFlexProcessing2(modelId) {
+  return modelId.startsWith("o3") || modelId.startsWith("o4-mini");
+}
 var openaiResponsesProviderOptionsSchema = import_zod12.z.object({
   metadata: import_zod12.z.any().nullish(),
   parallelToolCalls: import_zod12.z.boolean().nullish(),
@@ -2569,7 +2627,8 @@ var openaiResponsesProviderOptionsSchema = import_zod12.z.object({
   reasoningEffort: import_zod12.z.string().nullish(),
   strictSchemas: import_zod12.z.boolean().nullish(),
   instructions: import_zod12.z.string().nullish(),
-  reasoningSummary: import_zod12.z.string().nullish()
+  reasoningSummary: import_zod12.z.string().nullish(),
+  serviceTier: import_zod12.z.enum(["auto", "flex"]).nullish()
 });
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {