@ai-sdk/openai 4.0.56 → 4.0.58

package/CHANGELOG.md

@@ -1,5 +1,19 @@
 # @ai-sdk/openai
 
+## 4.0.58
+
+### Patch Changes
+
+- b54e551: fix(openai): support documented Responses image generation options
+- 17d3436: feat(openai): add support for ultrafast service tier
+
+## 4.0.57
+
+### Patch Changes
+
+- 048ce06: feat(batch): surface the uploaded input file on the batch start result (`providerMetadata.<provider>.inputFileId` / `inputFileExpiresAt`) and accept an `inputFileExpiresAfter` provider option on the OpenAI and xAI batch input file upload
+- 7243530: fix(openai): preserve complete Responses API raw usage objects
+
 ## 4.0.56
 
 ### Patch Changes

package/dist/index.d.ts

@@ -15,7 +15,7 @@ declare const openaiLanguageModelChatOptions: _ai_sdk_provider_utils.LazySchema<
     store?: boolean | undefined;
     metadata?: Record<string, string> | undefined;
     prediction?: Record<string, any> | undefined;
-    serviceTier?: "default" | "auto" | "flex" | "priority" | "fast" | undefined;
+    serviceTier?: "default" | "auto" | "flex" | "priority" | "fast" | "ultrafast" | undefined;
     strictJsonSchema?: boolean | undefined;
     textVerbosity?: "low" | "medium" | "high" | undefined;
     promptCacheKey?: string | undefined;
@@ -255,20 +255,21 @@ declare const openaiResponsesChunkSchema: _ai_sdk_provider_utils.LazySchema<{
         incomplete_details?: {
             reason: string;
         } | null | undefined;
-        usage?: {
+        usage?: (Record<string, JSONValue | undefined> & {
             input_tokens: number;
             output_tokens: number;
-            input_tokens_details?: {
+            input_tokens_details?: (Record<string, JSONValue | undefined> & {
                 cached_tokens?: number | null | undefined;
                 cache_write_tokens?: number | null | undefined;
                 orchestration_input_tokens?: number | null | undefined;
                 orchestration_input_cached_tokens?: number | null | undefined;
-            } | null | undefined;
-            output_tokens_details?: {
+            }) | null | undefined;
+            output_tokens_details?: (Record<string, JSONValue | undefined> & {
                 reasoning_tokens?: number | null | undefined;
                 orchestration_output_tokens?: number | null | undefined;
-            } | null | undefined;
-        } | null | undefined;
+            }) | null | undefined;
+            total_tokens?: number | undefined;
+        }) | null | undefined;
         reasoning?: {
             context?: string | null | undefined;
         } | null | undefined;
@@ -285,20 +286,21 @@ declare const openaiResponsesChunkSchema: _ai_sdk_provider_utils.LazySchema<{
         incomplete_details?: {
             reason: string;
         } | null | undefined;
-        usage?: {
+        usage?: (Record<string, JSONValue | undefined> & {
             input_tokens: number;
             output_tokens: number;
-            input_tokens_details?: {
+            input_tokens_details?: (Record<string, JSONValue | undefined> & {
                 cached_tokens?: number | null | undefined;
                 cache_write_tokens?: number | null | undefined;
                 orchestration_input_tokens?: number | null | undefined;
                 orchestration_input_cached_tokens?: number | null | undefined;
-            } | null | undefined;
-            output_tokens_details?: {
+            }) | null | undefined;
+            output_tokens_details?: (Record<string, JSONValue | undefined> & {
                 reasoning_tokens?: number | null | undefined;
                 orchestration_output_tokens?: number | null | undefined;
-            } | null | undefined;
-        } | null | undefined;
+            }) | null | undefined;
+            total_tokens?: number | undefined;
+        }) | null | undefined;
         reasoning?: {
             context?: string | null | undefined;
         } | null | undefined;
@@ -1126,6 +1128,7 @@ declare const openaiTools: {
      * @param size - The size of the generated image. One of 'auto', '1024x1024', '1024x1536', or '1536x1024'.
      */
     imageGeneration: (args?: {
+        action?: "generate" | "edit" | "auto";
         background?: "auto" | "opaque" | "transparent";
         inputFidelity?: "low" | "high";
         inputImageMask?: {
@@ -1133,12 +1136,12 @@ declare const openaiTools: {
             imageUrl?: string;
         };
         model?: string;
-        moderation?: "auto";
+        moderation?: "auto" | "low";
         outputCompression?: number;
         outputFormat?: "png" | "jpeg" | "webp";
         partialImages?: number;
         quality?: "auto" | "low" | "medium" | "high";
-        size?: "auto" | "1024x1024" | "1024x1536" | "1536x1024";
+        size?: "auto" | "1024x1024" | "1024x1536" | "1536x1024" | (string & {});
     }) => _ai_sdk_provider_utils.ProviderExecutedTool<{}, {
         result: string;
     }, {}>;
@@ -1383,7 +1386,7 @@ declare const openaiLanguageModelResponsesOptionsSchema: _ai_sdk_provider_utils.
     reasoningContext?: "auto" | "current_turn" | "all_turns" | undefined;
     reasoningSummary?: string | null | undefined;
     safetyIdentifier?: string | null | undefined;
-    serviceTier?: "default" | "auto" | "flex" | "priority" | "fast" | null | undefined;
+    serviceTier?: "default" | "auto" | "flex" | "priority" | "fast" | "ultrafast" | null | undefined;
     store?: boolean | null | undefined;
     passThroughUnsupportedFiles?: boolean | undefined;
     strictJsonSchema?: boolean | null | undefined;

package/dist/index.js

@@ -837,11 +837,12 @@ var openaiLanguageModelChatOptions = lazySchema2(
        * - 'flex': 50% cheaper processing at the cost of increased latency. Only available for o3 and o4-mini models.
        * - 'priority': Higher-speed processing with predictably low latency at premium cost. Available for Enterprise customers.
        * - 'fast': OpenAI's newer name for the 'priority' tier. Interchangeable with it.
+       * - 'ultrafast': Access-controlled Ultrafast processing. Only available for gpt-5.6-sol.
        * - 'default': The request will be processed with the standard pricing and performance for the selected model.
        *
        * @default 'auto'
        */
-      serviceTier: z3.enum(["auto", "flex", "priority", "fast", "default"]).optional(),
+      serviceTier: z3.enum(["auto", "flex", "priority", "fast", "ultrafast", "default"]).optional(),
       /**
        * Whether to use strict JSON schema validation.
        *
@@ -2957,6 +2958,7 @@ import { z as z17 } from "zod/v4";
 var imageGenerationArgsSchema = lazySchema16(
   () => zodSchema16(
     z17.object({
+      action: z17.enum(["generate", "edit", "auto"]).optional(),
       background: z17.enum(["auto", "opaque", "transparent"]).optional(),
       inputFidelity: z17.enum(["low", "high"]).optional(),
       inputImageMask: z17.object({
@@ -2964,12 +2966,15 @@ var imageGenerationArgsSchema = lazySchema16(
         imageUrl: z17.string().optional()
       }).optional(),
       model: z17.string().optional(),
-      moderation: z17.enum(["auto"]).optional(),
+      moderation: z17.enum(["auto", "low"]).optional(),
       outputCompression: z17.number().int().min(0).max(100).optional(),
       outputFormat: z17.enum(["png", "jpeg", "webp"]).optional(),
       partialImages: z17.number().int().min(0).max(3).optional(),
       quality: z17.enum(["auto", "low", "medium", "high"]).optional(),
-      size: z17.enum(["1024x1024", "1024x1536", "1536x1024", "auto"]).optional()
+      size: z17.union([
+        z17.enum(["1024x1024", "1024x1536", "1536x1024", "auto"]),
+        z17.string().regex(/^\d+x\d+$/)
+      ]).optional()
     }).strict()
   )
 );
@@ -3529,6 +3534,7 @@ import {
   getFromApi as getFromApi2,
   lazySchema as lazySchema26,
   normalizeBatchRequestCounts,
+  parseProviderOptions as parseProviderOptions8,
   postJsonToApi as postJsonToApi6,
   postToApi,
   safeValidateTypes,
@@ -3623,6 +3629,31 @@ var jsonValueSchema2 = z25.lazy(
     z25.record(z25.string(), jsonValueSchema2.optional())
   ])
 );
+var jsonObjectSchema = z25.record(z25.string(), jsonValueSchema2.optional());
+var openaiResponsesUsageSchema = z25.intersection(
+  jsonObjectSchema,
+  z25.object({
+    input_tokens: z25.number(),
+    input_tokens_details: z25.intersection(
+      jsonObjectSchema,
+      z25.object({
+        cached_tokens: z25.number().nullish(),
+        cache_write_tokens: z25.number().nullish(),
+        orchestration_input_tokens: z25.number().nullish(),
+        orchestration_input_cached_tokens: z25.number().nullish()
+      })
+    ).nullish(),
+    output_tokens: z25.number(),
+    output_tokens_details: z25.intersection(
+      jsonObjectSchema,
+      z25.object({
+        reasoning_tokens: z25.number().nullish(),
+        orchestration_output_tokens: z25.number().nullish()
+      })
+    ).nullish(),
+    total_tokens: z25.number().optional()
+  })
+);
 var openaiResponsesComputerSafetyCheckSchema = z25.object({
   id: z25.string(),
   code: z25.string().nullish(),
@@ -3819,20 +3850,7 @@ var openaiResponsesChunkSchema = lazySchema24(
         type: z25.enum(["response.completed", "response.incomplete"]),
         response: z25.object({
           incomplete_details: z25.object({ reason: z25.string() }).nullish(),
-          usage: z25.object({
-            input_tokens: z25.number(),
-            input_tokens_details: z25.object({
-              cached_tokens: z25.number().nullish(),
-              cache_write_tokens: z25.number().nullish(),
-              orchestration_input_tokens: z25.number().nullish(),
-              orchestration_input_cached_tokens: z25.number().nullish()
-            }).nullish(),
-            output_tokens: z25.number(),
-            output_tokens_details: z25.object({
-              reasoning_tokens: z25.number().nullish(),
-              orchestration_output_tokens: z25.number().nullish()
-            }).nullish()
-          }).nullish(),
+          usage: openaiResponsesUsageSchema.nullish(),
           reasoning: z25.object({
             context: z25.string().nullish()
           }).nullish(),
@@ -3848,20 +3866,7 @@ var openaiResponsesChunkSchema = lazySchema24(
             message: z25.string()
           }).nullish(),
           incomplete_details: z25.object({ reason: z25.string() }).nullish(),
-          usage: z25.object({
-            input_tokens: z25.number(),
-            input_tokens_details: z25.object({
-              cached_tokens: z25.number().nullish(),
-              cache_write_tokens: z25.number().nullish(),
-              orchestration_input_tokens: z25.number().nullish(),
-              orchestration_input_cached_tokens: z25.number().nullish()
-            }).nullish(),
-            output_tokens: z25.number(),
-            output_tokens_details: z25.object({
-              reasoning_tokens: z25.number().nullish(),
-              orchestration_output_tokens: z25.number().nullish()
-            }).nullish()
-          }).nullish(),
+          usage: openaiResponsesUsageSchema.nullish(),
           reasoning: z25.object({
             context: z25.string().nullish()
           }).nullish(),
@@ -4651,20 +4656,7 @@ var openaiResponsesResponseSchema = lazySchema24(
         context: z25.string().nullish()
       }).nullish(),
       incomplete_details: z25.object({ reason: z25.string() }).nullish(),
-      usage: z25.object({
-        input_tokens: z25.number(),
-        input_tokens_details: z25.object({
-          cached_tokens: z25.number().nullish(),
-          cache_write_tokens: z25.number().nullish(),
-          orchestration_input_tokens: z25.number().nullish(),
-          orchestration_input_cached_tokens: z25.number().nullish()
-        }).nullish(),
-        output_tokens: z25.number(),
-        output_tokens_details: z25.object({
-          reasoning_tokens: z25.number().nullish(),
-          orchestration_output_tokens: z25.number().nullish()
-        }).nullish()
-      }).nullish()
+      usage: openaiResponsesUsageSchema.nullish()
     })
   )
 );
@@ -6106,10 +6098,11 @@ var openaiLanguageModelResponsesOptionsSchema = lazySchema25(
        * Set to 'flex' for 50% cheaper processing at the cost of increased latency (available for o3, o4-mini, and gpt-5 models).
        * Set to 'priority' for faster processing with Enterprise access (available for gpt-4, gpt-5, gpt-5-mini, o3, o4-mini; gpt-5-nano is not supported).
        * Set to 'fast' for the same tier as 'priority' (OpenAI's newer name for it).
+       * Set to 'ultrafast' for access-controlled Ultrafast processing (available only for gpt-5.6-sol).
        *
        * Defaults to 'auto'.
        */
-      serviceTier: z27.enum(["auto", "flex", "priority", "fast", "default"]).nullish(),
+      serviceTier: z27.enum(["auto", "flex", "priority", "fast", "ultrafast", "default"]).nullish(),
       /**
        * Whether to store the generation. Defaults to `true`.
        */
@@ -6381,6 +6374,7 @@ async function prepareResponsesTools({
             });
             openaiTools2.push({
               type: "image_generation",
+              action: args.action,
               background: args.background,
               input_fidelity: args.inputFidelity,
               input_image_mask: args.inputImageMask ? {
@@ -8924,7 +8918,19 @@ function escapeJSONDelta(delta) {
 
 // src/openai-responses-batch.ts
 var openaiBatchEndpoint = "/v1/responses";
-var openaiBatchInputFileExpiresAfterSeconds = 48 * 60 * 60;
+var openaiBatchInputFileDefaultExpiresAfterSeconds = 48 * 60 * 60;
+var openaiBatchProviderOptionsSchema = lazySchema26(
+  () => zodSchema26(
+    z28.object({
+      /**
+       * TTL in seconds for the uploaded batch input file, measured from
+       * upload time. OpenAI accepts integers between 3600 (1 hour) and
+       * 2592000 (30 days) inclusive. Defaults to 48 hours.
+       */
+      inputFileExpiresAfter: z28.number().int().min(3600).max(2592e3).optional()
+    })
+  )
+);
 var openaiBatchResponseSchema = lazySchema26(
   () => zodSchema26(
     z28.object({
@@ -8971,7 +8977,7 @@ var OpenAIResponsesBatch = class {
     this.options = options;
   }
   async startBatch(options) {
-    var _a2, _b, _c, _d, _e;
+    var _a2, _b, _c, _d, _e, _f;
     const fileParts = [];
     const warnings = options.webhookUrl == null ? [] : [
       {
@@ -8982,6 +8988,10 @@ var OpenAIResponsesBatch = class {
         }
       }
     ];
+    const batchOptions = await this.parseBatchProviderOptions(
+      options.providerOptions
+    );
+    const inputFileExpiresAfterSeconds = (_a2 = batchOptions == null ? void 0 : batchOptions.inputFileExpiresAfter) != null ? _a2 : openaiBatchInputFileDefaultExpiresAfterSeconds;
     for (const request of options.requests) {
       const preparedRequest = await this.options.prepareRequest(request);
       fileParts.push(
@@ -8996,7 +9006,7 @@ var OpenAIResponsesBatch = class {
       for (const warning of preparedRequest.warnings) {
         warnings.push({ requestId: request.id, warning });
       }
-      for (const tool of (_a2 = request.options.tools) != null ? _a2 : []) {
+      for (const tool of (_b = request.options.tools) != null ? _b : []) {
         if (tool.type === "provider" && !openAIBatchConvertibleProviderToolIds.has(tool.id)) {
           warnings.push({
             requestId: request.id,
@@ -9020,19 +9030,17 @@ var OpenAIResponsesBatch = class {
     formData.append("expires_after[anchor]", "created_at");
     formData.append(
       "expires_after[seconds]",
-      String(openaiBatchInputFileExpiresAfterSeconds)
+      String(inputFileExpiresAfterSeconds)
     );
     const { value: uploadedFile } = await postToApi({
       url: this.getUrl("/files"),
-      headers: combineHeaders7((_c = (_b = this.options.config).headers) == null ? void 0 : _c.call(_b), options.headers),
+      headers: combineHeaders7((_d = (_c = this.options.config).headers) == null ? void 0 : _d.call(_c), options.headers),
       body: {
         content: formData,
         values: {
           purpose: "batch",
           "expires_after[anchor]": "created_at",
-          "expires_after[seconds]": String(
-            openaiBatchInputFileExpiresAfterSeconds
-          ),
+          "expires_after[seconds]": String(inputFileExpiresAfterSeconds),
           file: {
             name: filename,
             type: file.type,
@@ -9049,7 +9057,7 @@ var OpenAIResponsesBatch = class {
     });
     const { value: batch } = await postJsonToApi6({
       url: this.getUrl("/batches"),
-      headers: combineHeaders7((_e = (_d = this.options.config).headers) == null ? void 0 : _e.call(_d), options.headers),
+      headers: combineHeaders7((_f = (_e = this.options.config).headers) == null ? void 0 : _f.call(_e), options.headers),
       body: {
         input_file_id: uploadedFile.id,
         endpoint: openaiBatchEndpoint,
@@ -9062,12 +9070,35 @@ var OpenAIResponsesBatch = class {
       abortSignal: options.abortSignal,
       fetch: this.options.config.fetch
     });
+    const inputFileExpiresAt = convertUnixTimestamp(uploadedFile.expires_at);
     return {
       batchId: batch.id,
       ...convertOpenAIBatchStatus(batch),
+      providerMetadata: {
+        openai: {
+          inputFileId: uploadedFile.id,
+          ...inputFileExpiresAt != null ? { inputFileExpiresAt } : {}
+        }
+      },
       warnings
     };
   }
+  async parseBatchProviderOptions(providerOptions) {
+    const providerOptionsName = this.options.config.provider.includes("azure") ? "azure" : "openai";
+    let batchOptions = await parseProviderOptions8({
+      provider: providerOptionsName,
+      providerOptions,
+      schema: openaiBatchProviderOptionsSchema
+    });
+    if (batchOptions == null && providerOptionsName !== "openai") {
+      batchOptions = await parseProviderOptions8({
+        provider: "openai",
+        providerOptions,
+        schema: openaiBatchProviderOptionsSchema
+      });
+    }
+    return batchOptions;
+  }
   async getBatchStatus(options) {
     const batch = await this.retrieveBatch(options);
     return convertOpenAIBatchStatus(batch);
@@ -9896,7 +9927,7 @@ var OpenAIRealtimeModel = class {
 import {
   combineHeaders as combineHeaders8,
   createBinaryResponseHandler,
-  parseProviderOptions as parseProviderOptions8,
+  parseProviderOptions as parseProviderOptions9,
   postJsonToApi as postJsonToApi7,
   serializeModelOptions as serializeModelOptions6,
   WORKFLOW_DESERIALIZE as WORKFLOW_DESERIALIZE7,
@@ -9947,7 +9978,7 @@ var OpenAISpeechModel = class _OpenAISpeechModel {
     providerOptions
   }) {
     const warnings = [];
-    const openAIOptions = await parseProviderOptions8({
+    const openAIOptions = await parseProviderOptions9({
       provider: "openai",
       providerOptions,
       schema: openaiSpeechModelOptionsSchema
@@ -10039,7 +10070,7 @@ import {
   createJsonResponseHandler as createJsonResponseHandler8,
   connectToWebSocket,
   mediaTypeToExtension,
-  parseProviderOptions as parseProviderOptions9,
+  parseProviderOptions as parseProviderOptions10,
   postFormDataToApi as postFormDataToApi3,
   safeParseJSON as safeParseJSON2,
   serializeModelOptions as serializeModelOptions7,
@@ -10218,7 +10249,7 @@ var OpenAITranscriptionModel = class _OpenAITranscriptionModel {
     providerOptions
   }) {
     const warnings = [];
-    const openAIOptions = await parseProviderOptions9({
+    const openAIOptions = await parseProviderOptions10({
       provider: "openai",
       providerOptions,
       schema: openAITranscriptionModelOptions
@@ -10327,7 +10358,7 @@ var OpenAITranscriptionModel = class _OpenAITranscriptionModel {
       });
     }
     const currentDate = (_c = (_b = (_a2 = this.config._internal) == null ? void 0 : _a2.currentDate) == null ? void 0 : _b.call(_a2)) != null ? _c : /* @__PURE__ */ new Date();
-    const openAIOptions = await parseProviderOptions9({
+    const openAIOptions = await parseProviderOptions10({
       provider: "openai",
       providerOptions: options.providerOptions,
       schema: openAITranscriptionModelOptions
@@ -10569,7 +10600,7 @@ import {
   combineHeaders as combineHeaders10,
   connectToWebSocket as connectToWebSocket2,
   convertToBase64 as convertToBase644,
-  parseProviderOptions as parseProviderOptions10,
+  parseProviderOptions as parseProviderOptions11,
   safeParseJSON as safeParseJSON3,
   serializeModelOptions as serializeModelOptions8,
   toWebSocketUrl as toWebSocketUrl2,
@@ -10616,7 +10647,7 @@ var OpenAISpeechTranslationModel = class _OpenAISpeechTranslationModel {
       });
     }
     const currentDate = (_c = (_b = (_a2 = this.config._internal) == null ? void 0 : _a2.currentDate) == null ? void 0 : _b.call(_a2)) != null ? _c : /* @__PURE__ */ new Date();
-    await parseProviderOptions10({
+    await parseProviderOptions11({
       provider: "openai",
       providerOptions: options.providerOptions,
       schema: openAISpeechTranslationModelOptions
@@ -10963,7 +10994,7 @@ var OpenAISkills = class {
 };
 
 // src/version.ts
-var VERSION = true ? "4.0.56" : "0.0.0-test";
+var VERSION = true ? "4.0.58" : "0.0.0-test";
 
 // src/openai-provider.ts
 function createOpenAI(options = {}) {