@ai-sdk/openai 4.0.58 → 4.0.59

package/docs/03-openai.mdx

@@ -2720,6 +2720,7 @@ The following optional provider options are available for OpenAI completion mode
 
 | Model                 | Image Input | Audio Input | Object Generation | Tool Usage |
 | --------------------- | ----------- | ----------- | ----------------- | ---------- |
+| `gpt-6-astra`         | <Check />   | <Cross />   | <Check />         | <Check />  |
 | `gpt-5.6`             | <Check />   | <Cross />   | <Check />         | <Check />  |
 | `gpt-5.6-luna`        | <Check />   | <Cross />   | <Check />         | <Check />  |
 | `gpt-5.6-sol`         | <Check />   | <Cross />   | <Check />         | <Check />  |
@@ -3022,6 +3023,21 @@ const result = await transcribe({
 console.log(result.segments); // Array of segments with startSecond/endSecond
 ```
 
+`gpt-4o-transcribe-diarize` identifies speakers in a recording. It defaults to
+the `diarized_json` response format and automatic chunking:
+
+```ts
+import { transcribe } from 'ai';
+import { openai } from '@ai-sdk/openai';
+
+const result = await transcribe({
+  model: openai.transcription('gpt-4o-transcribe-diarize'),
+  audio: new Uint8Array([1, 2, 3, 4]),
+});
+
+console.log(result.providerMetadata.openai.segments);
+```
+
 The following provider options are available:
 
 - **timestampGranularities** _string[]_
@@ -3046,6 +3062,12 @@ The following provider options are available:
 - **include** _string[]_
   Additional information to include in the transcription response.
 
+- **responseFormat** _'json' | 'verbose_json' | 'diarized_json'_
+  The format of the transcription response. `gpt-4o-transcribe-diarize` defaults to `diarized_json`.
+
+- **chunkingStrategy** _'auto' | object_
+  Controls how the audio is split into chunks. `gpt-4o-transcribe-diarize` defaults to `'auto'`; provide this option to override the default. The object form configures OpenAI server-side VAD with `type: 'server_vad'` and optional `threshold`, `prefixPaddingMs`, and `silenceDurationMs` fields.
+
 - **streaming** _object_
   Options for streaming transcription models such as `gpt-realtime-whisper`.
   Use with `experimental_streamTranscribe`.
@@ -3057,12 +3079,13 @@ The following provider options are available:
 
 ### Model Capabilities
 
-| Model                    | Transcription | Streaming | Duration  | Segments  | Language  |
-| ------------------------ | ------------- | --------- | --------- | --------- | --------- |
-| `whisper-1`              | <Check />     | <Cross /> | <Check /> | <Check /> | <Check /> |
-| `gpt-4o-mini-transcribe` | <Check />     | <Cross /> | <Cross /> | <Cross /> | <Cross /> |
-| `gpt-4o-transcribe`      | <Check />     | <Cross /> | <Cross /> | <Cross /> | <Cross /> |
-| `gpt-realtime-whisper`   | <Cross />     | <Check /> | <Cross /> | <Cross /> | <Cross /> |
+| Model                       | Transcription | Streaming | Duration  | Segments  | Language  |
+| --------------------------- | ------------- | --------- | --------- | --------- | --------- |
+| `whisper-1`                 | <Check />     | <Cross /> | <Check /> | <Check /> | <Check /> |
+| `gpt-4o-mini-transcribe`    | <Check />     | <Cross /> | <Cross /> | <Cross /> | <Cross /> |
+| `gpt-4o-transcribe`         | <Check />     | <Cross /> | <Cross /> | <Cross /> | <Cross /> |
+| `gpt-4o-transcribe-diarize` | <Check />     | <Cross /> | <Check /> | <Check /> | <Cross /> |
+| `gpt-realtime-whisper`      | <Cross />     | <Check /> | <Cross /> | <Cross /> | <Cross /> |
 
 ## Translation Models
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/openai",
-  "version": "4.0.58",
+  "version": "4.0.59",
   "type": "module",
   "license": "Apache-2.0",
   "sideEffects": false,

package/src/chat/openai-chat-language-model-options.ts

@@ -70,6 +70,7 @@ export type OpenAIChatModelId =
   | 'gpt-5.6-luna'
   | 'gpt-5.6-sol'
   | 'gpt-5.6-terra'
+  | 'gpt-6-astra'
   | (string & {});
 
 export const openaiLanguageModelChatOptions = lazySchema(() =>

package/src/responses/openai-responses-language-model-options.ts

@@ -57,6 +57,7 @@ export const openaiResponsesReasoningModelIds = [
   'gpt-5.6-luna',
   'gpt-5.6-sol',
   'gpt-5.6-terra',
+  'gpt-6-astra',
 ] as const;
 
 export const openaiResponsesModelIds = [
@@ -129,6 +130,7 @@ export type OpenAIResponsesModelId =
   | 'gpt-5.6-luna'
   | 'gpt-5.6-sol'
   | 'gpt-5.6-terra'
+  | 'gpt-6-astra'
   | 'gpt-5-2025-08-07'
   | 'gpt-5-chat-latest'
   | 'gpt-5-codex'

package/src/transcription/openai-transcription-api.ts

@@ -18,18 +18,28 @@ export const openaiTranscriptionResponseSchema = lazySchema(() =>
         .nullish(),
       segments: z
         .array(
-          z.object({
-            id: z.number(),
-            seek: z.number(),
-            start: z.number(),
-            end: z.number(),
-            text: z.string(),
-            tokens: z.array(z.number()),
-            temperature: z.number(),
-            avg_logprob: z.number(),
-            compression_ratio: z.number(),
-            no_speech_prob: z.number(),
-          }),
+          z.union([
+            z.object({
+              id: z.number(),
+              seek: z.number(),
+              start: z.number(),
+              end: z.number(),
+              text: z.string(),
+              tokens: z.array(z.number()),
+              temperature: z.number(),
+              avg_logprob: z.number(),
+              compression_ratio: z.number(),
+              no_speech_prob: z.number(),
+            }),
+            z.object({
+              type: z.literal('transcript.text.segment'),
+              id: z.string(),
+              start: z.number(),
+              end: z.number(),
+              text: z.string(),
+              speaker: z.string(),
+            }),
+          ]),
         )
         .nullish(),
     }),

package/src/transcription/openai-transcription-model-options.ts

@@ -50,6 +50,28 @@ export const openAITranscriptionModelOptions = lazySchema(() =>
         .default(['segment'])
         .optional(),
 
+      /**
+       * The format of the transcription response.
+       */
+      responseFormat: z
+        .enum(['json', 'verbose_json', 'diarized_json'])
+        .optional(),
+
+      /**
+       * Controls how the audio is split into chunks before transcription.
+       */
+      chunkingStrategy: z
+        .union([
+          z.literal('auto'),
+          z.object({
+            type: z.literal('server_vad'),
+            threshold: z.number().min(0).max(1).optional(),
+            prefixPaddingMs: z.number().int().min(0).optional(),
+            silenceDurationMs: z.number().int().min(0).optional(),
+          }),
+        ])
+        .optional(),
+
       /**
        * Options for streaming transcription models such as `gpt-realtime-whisper`.
        */

package/src/transcription/openai-transcription-model.ts

@@ -195,6 +195,11 @@ export class OpenAITranscriptionModel implements TranscriptionModelV4 {
       formData.append('response_format', 'verbose_json');
     }
 
+    const isDiarizationModel = this.modelId === 'gpt-4o-transcribe-diarize';
+    const chunkingStrategy =
+      openAIOptions?.chunkingStrategy ??
+      (isDiarizationModel ? 'auto' : undefined);
+
     // Add provider-specific options
     if (openAIOptions) {
       const isGpt4oTranscribeModel = [
@@ -209,7 +214,13 @@ export class OpenAITranscriptionModel implements TranscriptionModelV4 {
         // https://platform.openai.com/docs/api-reference/audio/createTranscription#audio_createtranscription-response_format
         // prefer verbose_json to get segments for models that support it
         ...(this.modelId !== 'whisper-1' && {
-          response_format: isGpt4oTranscribeModel ? 'json' : 'verbose_json',
+          response_format:
+            openAIOptions.responseFormat ??
+            (isDiarizationModel
+              ? 'diarized_json'
+              : isGpt4oTranscribeModel
+                ? 'json'
+                : 'verbose_json'),
         }),
         temperature: openAIOptions.temperature,
         timestamp_granularities: openAIOptions.timestampGranularities,
@@ -226,6 +237,28 @@ export class OpenAITranscriptionModel implements TranscriptionModelV4 {
           }
         }
       }
+    } else if (isDiarizationModel) {
+      formData.append('response_format', 'diarized_json');
+    }
+
+    if (chunkingStrategy != null) {
+      formData.append(
+        'chunking_strategy',
+        typeof chunkingStrategy === 'string'
+          ? chunkingStrategy
+          : JSON.stringify({
+              type: chunkingStrategy.type,
+              ...(chunkingStrategy.threshold != null && {
+                threshold: chunkingStrategy.threshold,
+              }),
+              ...(chunkingStrategy.prefixPaddingMs != null && {
+                prefix_padding_ms: chunkingStrategy.prefixPaddingMs,
+              }),
+              ...(chunkingStrategy.silenceDurationMs != null && {
+                silence_duration_ms: chunkingStrategy.silenceDurationMs,
+              }),
+            }),
+      );
     }
 
     return {
@@ -270,6 +303,19 @@ export class OpenAITranscriptionModel implements TranscriptionModelV4 {
         ? languageMap[response.language as keyof typeof languageMap]
         : undefined;
 
+    const diarizedSegments = response.segments?.flatMap(segment =>
+      'speaker' in segment
+        ? [
+            {
+              text: segment.text,
+              startSecond: segment.start,
+              endSecond: segment.end,
+              speaker: segment.speaker,
+            },
+          ]
+        : [],
+    );
+
     return {
       text: response.text,
       segments:
@@ -293,6 +339,14 @@ export class OpenAITranscriptionModel implements TranscriptionModelV4 {
         headers: responseHeaders,
         body: rawResponse,
       },
+      ...(diarizedSegments != null &&
+        diarizedSegments.length > 0 && {
+          providerMetadata: {
+            openai: {
+              segments: diarizedSegments,
+            },
+          },
+        }),
     };
   }