@@ -0,0 +1,7 @@
+---
+'@ai-sdk/provider': patch
+'@ai-sdk/openai': patch
+'ai': patch
+---
+
+core (ai): change transcription model mimeType to mediaType
@@ -0,0 +1,5 @@
+---
+'@ai-sdk/openai': patch
+---
+
+fix (provider/openai): increase transcription model resilience
@@ -6,9 +6,7 @@ import { readFile } from 'fs/promises';
 async function main() {
   const result = await transcribe({
     model: openai.transcription('whisper-1'),
-    audio: Buffer.from(
-      await readFile('examples/ai-core/data/galileo.mp3'),
-    ).toString('base64'),
+    audio: Buffer.from(await readFile('./data/galileo.mp3')).toString('base64'),
   });
 
   console.log('Text:', result.text);
 
@@ -6,8 +6,7 @@ async function main() {
   const result = await transcribe({
     model: openai.transcription('whisper-1'),
     audio: new URL(
-      '/vercel/ai/raw/refs/heads/main/examples/ai-core/data/galileo.mp3',
-      'https://github.com',
+      'https://github.com/vercel/ai/raw/refs/heads/main/examples/ai-core/data/galileo.mp3',
     ),
   });
 
 
@@ -6,7 +6,7 @@ import { readFile } from 'fs/promises';
 async function main() {
   const result = await transcribe({
     model: openai.transcription('whisper-1'),
-    audio: await readFile('examples/ai-core/data/galileo.mp3'),
+    audio: await readFile('data/galileo.mp3'),
   });
 
   console.log('Text:', result.text);
 
@@ -78,7 +78,7 @@ describe('transcribe', () => {
 
     expect(capturedArgs).toStrictEqual({
       audio: audioData,
-      mimeType: 'audio/wav',
+      mediaType: 'audio/wav',
       headers: { 'custom-request-header': 'request-header-value' },
       abortSignal,
       providerOptions: {},
 
@@ -1,17 +1,17 @@
-import { TranscriptionModelV1, JSONValue } from '@ai-sdk/provider';
+import { JSONValue, TranscriptionModelV1 } from '@ai-sdk/provider';
 import { NoTranscriptGeneratedError } from '../../errors/no-transcript-generated-error';
+import { download } from '../../util/download';
+import { DataContent } from '../prompt';
+import { convertDataContentToUint8Array } from '../prompt/data-content';
 import { prepareRetries } from '../prompt/prepare-retries';
+import { ProviderOptions } from '../types/provider-metadata';
 import { TranscriptionWarning } from '../types/transcription-model';
 import { TranscriptionModelResponseMetadata } from '../types/transcription-model-response-metadata';
-import { TranscriptionResult } from './transcribe-result';
-import { DataContent } from '../prompt';
-import { convertDataContentToUint8Array } from '../prompt/data-content';
-import { download } from '../../util/download';
 import {
   audioMimeTypeSignatures,
   detectMimeType,
 } from '../util/detect-mimetype';
-import { ProviderOptions } from '../types/provider-metadata';
+import { TranscriptionResult } from './transcribe-result';
 
 /**
 Generates transcripts using a transcription model.
@@ -81,7 +81,7 @@ Only applicable for HTTP-based providers.
   const { retry } = prepareRetries({ maxRetries: maxRetriesArg });
   const audioData =
     audio instanceof URL
-      ? new Uint8Array((await download({ url: audio })).data)
+      ? (await download({ url: audio })).data
       : convertDataContentToUint8Array(audio);
 
   const result = await retry(() =>
@@ -90,7 +90,7 @@ Only applicable for HTTP-based providers.
       abortSignal,
       headers,
       providerOptions,
-      mimeType:
+      mediaType:
         detectMimeType({
           data: audioData,
           signatures: audioMimeTypeSignatures,
 
@@ -73,7 +73,7 @@ describe('doGenerate', () => {
 
     await model.doGenerate({
       audio: audioData,
-      mimeType: 'audio/wav',
+      mediaType: 'audio/wav',
     });
 
     expect(await server.calls[0].requestBodyMultipart).toMatchObject({
@@ -95,7 +95,7 @@ describe('doGenerate', () => {
 
     await provider.transcription('whisper-1').doGenerate({
       audio: audioData,
-      mimeType: 'audio/wav',
+      mediaType: 'audio/wav',
       headers: {
         'Custom-Request-Header': 'request-header-value',
       },
@@ -118,7 +118,7 @@ describe('doGenerate', () => {
 
     const result = await model.doGenerate({
       audio: audioData,
-      mimeType: 'audio/wav',
+      mediaType: 'audio/wav',
     });
 
     expect(result.text).toBe('Hello from the Vercel AI SDK!');
@@ -144,7 +144,7 @@ describe('doGenerate', () => {
 
     const result = await customModel.doGenerate({
       audio: audioData,
-      mimeType: 'audio/wav',
+      mediaType: 'audio/wav',
     });
 
     expect(result.response).toMatchObject({
@@ -173,10 +173,59 @@ describe('doGenerate', () => {
 
     const result = await customModel.doGenerate({
       audio: audioData,
-      mimeType: 'audio/wav',
+      mediaType: 'audio/wav',
     });
 
     expect(result.response.timestamp.getTime()).toEqual(testDate.getTime());
     expect(result.response.modelId).toBe('whisper-1');
   });
+
+  it('should work when no words, language, or duration are returned', async () => {
+    server.urls['https://api.openai.com/v1/audio/transcriptions'].response = {
+      type: 'json-value',
+      body: {
+        task: 'transcribe',
+        text: 'Hello from the Vercel AI SDK!',
+        _request_id: 'req_1234',
+      },
+    };
+
+    const testDate = new Date(0);
+    const customModel = new OpenAITranscriptionModel('whisper-1', {
+      provider: 'test-provider',
+      url: () => 'https://api.openai.com/v1/audio/transcriptions',
+      headers: () => ({}),
+      _internal: {
+        currentDate: () => testDate,
+      },
+    });
+
+    const result = await customModel.doGenerate({
+      audio: audioData,
+      mediaType: 'audio/wav',
+    });
+
+    expect(result).toMatchInlineSnapshot(`
+      {
+        "durationInSeconds": undefined,
+        "language": undefined,
+        "response": {
+          "body": {
+            "_request_id": "req_1234",
+            "task": "transcribe",
+            "text": "Hello from the Vercel AI SDK!",
+          },
+          "headers": {
+            "content-length": "85",
+            "content-type": "application/json",
+          },
+          "modelId": "whisper-1",
+          "timestamp": 1970-01-01T00:00:00.000Z,
+        },
+        "segments": [],
+        "text": "Hello from the Vercel AI SDK!",
+        "warnings": [],
+      }
+    `);
+  });
 });
@@ -142,7 +142,7 @@ export class OpenAITranscriptionModel implements TranscriptionModelV1 {
 
   private getArgs({
     audio,
-    mimeType,
+    mediaType,
     providerOptions,
   }: OpenAITranscriptionCallOptions) {
     const warnings: TranscriptionModelV1CallWarning[] = [];
@@ -162,7 +162,7 @@ export class OpenAITranscriptionModel implements TranscriptionModelV1 {
         : new Blob([convertBase64ToUint8Array(audio)]);
 
     formData.append('model', this.modelId);
-    formData.append('file', new File([blob], 'audio', { type: mimeType }));
+    formData.append('file', new File([blob], 'audio', { type: mediaType }));
 
     // Add provider-specific options
     if (openAIOptions) {
@@ -197,7 +197,11 @@ export class OpenAITranscriptionModel implements TranscriptionModelV1 {
     const currentDate = this.config._internal?.currentDate?.() ?? new Date();
     const { formData, warnings } = this.getArgs(options);
 
-    const { value: response, responseHeaders } = await postFormDataToApi({
+    const {
+      value: response,
+      responseHeaders,
+      rawValue: rawResponse,
+    } = await postFormDataToApi({
       url: this.config.url({
         path: '/audio/transcriptions',
         modelId: this.modelId,
@@ -212,48 +216,43 @@ export class OpenAITranscriptionModel implements TranscriptionModelV1 {
       fetch: this.config.fetch,
     });
 
-    let language: string | undefined;
-
-    if (response.language && response.language in languageMap) {
-      language = languageMap[response.language as keyof typeof languageMap];
-    }
+    const language =
+      response.language != null && response.language in languageMap
+        ? languageMap[response.language as keyof typeof languageMap]
+        : undefined;
 
     return {
       text: response.text,
-      segments: response.words.map(word => ({
-        text: word.word,
-        startSecond: word.start,
-        endSecond: word.end,
-      })),
+      segments:
+        response.words?.map(word => ({
+          text: word.word,
+          startSecond: word.start,
+          endSecond: word.end,
+        })) ?? [],
       language,
-      durationInSeconds: response.duration,
+      durationInSeconds: response.duration ?? undefined,
       warnings,
       response: {
         timestamp: currentDate,
         modelId: this.modelId,
         headers: responseHeaders,
-        body: response,
-      },
-
-      // When using format `verbose_json` on `whisper-1`, OpenAI includes the things like `task` and enhanced `segments` information.
-      providerMetadata: {
-        openai: {
-          transcript: response,
-        },
+        body: rawResponse,
       },
     };
   }
 }
 
 const openaiTranscriptionResponseSchema = z.object({
   text: z.string(),
-  language: z.string().optional(),
-  duration: z.number().optional(),
-  words: z.array(
-    z.object({
-      word: z.string(),
-      start: z.number(),
-      end: z.number(),
-    }),
-  ),
+  language: z.string().nullish(),
+  duration: z.number().nullish(),
+  words: z
+    .array(
+      z.object({
+        word: z.string(),
+        start: z.number(),
+        end: z.number(),
+      }),
+    )
+    .nullish(),
 });
@@ -13,9 +13,11 @@ Accepts a `Uint8Array` or `string`, where `string` is a base64 encoded audio fil
   audio: Uint8Array | string;
 
   /**
-  The MIME type of the audio data.
+The IANA media type of the audio data.
+
+@see https://www.iana.org/assignments/media-types/media-types.xhtml
    */
-  mimeType: string;
+  mediaType: string;
 
   /**
 Additional provider-specific options that are passed through to the provider
-Original file line number
+Diff line change
@@ @@ -0,0 +1,5 @@ @@
 +---
 +'@ai-sdk/openai': patch
 +---
++
 +fix (provider/openai): increase transcription model resilience