@ai-sdk/openai 4.0.103 → 4.0.104

package/docs/03-openai.mdx

@@ -245,9 +245,9 @@ The following provider options are available:
   Reasoning effort for reasoning models. Defaults to `medium`. If you use `providerOptions` to set the `reasoningEffort` option, this model setting will be ignored.
 
 <Note>
-  Supported reasoning efforts vary by model. GPT-5.6, GPT-6 Sol, and GPT-6 Luna
-  support `'none'`, `'low'`, `'medium'`, `'high'`, `'xhigh'`, and `'max'`. GPT-6
-  Astra and GPT-6.1 Sol do not support `'none'`.
+  Supported reasoning efforts vary by model. GPT-5.6, GPT-6 Sol, GPT-6 Luna, and
+  GPT-6.1 Terra support `'none'`, `'low'`, `'medium'`, `'high'`, `'xhigh'`, and
+  `'max'`. GPT-6 Astra and GPT-6.1 Sol do not support `'none'`.
 </Note>
 
 <Note type="warning">
@@ -3069,6 +3069,7 @@ The following optional provider options are available for OpenAI completion mode
 | --------------------- | ----------- | ----------- | ----------------- | ---------- |
 | `gpt-6-astra`         | <Check />   | <Cross />   | <Check />         | <Check />  |
 | `gpt-6.1-sol`         | <Check />   | <Cross />   | <Check />         | <Check />  |
+| `gpt-6.1-terra`       | <Check />   | <Cross />   | <Check />         | <Check />  |
 | `gpt-6-luna`          | <Check />   | <Cross />   | <Check />         | <Check />  |
 | `gpt-6-sol`           | <Check />   | <Cross />   | <Check />         | <Check />  |
 | `gpt-5.6`             | <Check />   | <Cross />   | <Check />         | <Check />  |
@@ -3508,12 +3509,13 @@ The following provider options are available:
 
 - **timestampGranularities** _string[]_
   The granularity of the timestamps in the transcription.
-  Defaults to `['segment']`.
+  Defaults to `['segment']` when `responseFormat` is `verbose_json`.
   Possible values are `['word']`, `['segment']`, and `['word', 'segment']`.
   Note: There is no additional latency for segment timestamps, but generating word timestamps incurs additional latency.
 
 - **language** _string_
   The language of the input audio. Supplying the input language in ISO-639-1 format (e.g. 'en') will improve accuracy and latency.
+  For streaming with `gpt-transcribe` and `gpt-live-transcribe`, it is sent as a single-entry `languages` list.
   Optional.
 
 - **prompt** _string_
@@ -3529,13 +3531,13 @@ The following provider options are available:
   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`.
+  The format of the transcription response. `whisper-1` defaults to `verbose_json` and `gpt-4o-transcribe-diarize` defaults to `diarized_json`. Other models use the service default unless this option is set.
 
 - **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`.
+  Options for streaming transcription models such as `gpt-live-transcribe`.
   Use with `experimental_streamTranscribe`.
   - **delay** _'minimal' | 'low' | 'medium' | 'high' | 'xhigh'_
     Latency/accuracy tradeoff for realtime transcription.
@@ -3551,6 +3553,8 @@ The following provider options are available:
 | `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-transcribe`            | <Check />     | <Check /> | <Cross /> | <Cross /> | <Check /> |
+| `gpt-live-transcribe`       | <Cross />     | <Check /> | <Cross /> | <Cross /> | <Cross /> |
 | `gpt-realtime-whisper`      | <Cross />     | <Check /> | <Cross /> | <Cross /> | <Cross /> |
 
 ## Translation Models

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/openai",
-  "version": "4.0.103",
+  "version": "4.0.104",
   "type": "module",
   "license": "Apache-2.0",
   "sideEffects": false,
@@ -36,7 +36,7 @@
   },
   "dependencies": {
     "@ai-sdk/provider": "4.0.34",
-    "@ai-sdk/provider-utils": "5.0.69"
+    "@ai-sdk/provider-utils": "5.0.70"
   },
   "devDependencies": {
     "@ai-sdk/test-server": "2.0.4",

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

@@ -72,6 +72,7 @@ export type OpenAIChatModelId =
   | 'gpt-5.6-terra'
   | 'gpt-6-astra'
   | 'gpt-6.1-sol'
+  | 'gpt-6.1-terra'
   | 'gpt-6-luna'
   | 'gpt-6-sol'
   | (string & {});

package/src/normalize-openai-json-schema.ts

@@ -58,9 +58,10 @@ export function normalizeOpenAIJsonSchema(schema: JSONSchema7): {
     const propertyNames = schema.propertyNames;
 
     if (propertyNames != null) {
+      // Property names are always strings in JSON, so omitted type is a string schema.
       if (
         typeof propertyNames === 'boolean' ||
-        propertyNames.type !== 'string'
+        (propertyNames.type != null && propertyNames.type !== 'string')
       ) {
         throw new UnsupportedFunctionalityError({
           functionality:

package/src/openai-batch.ts

@@ -47,7 +47,7 @@ import {
   type OpenAIResponsesLogprobs,
 } from './responses/openai-responses-api';
 import {
-  mapWebSearchOutput,
+  mapWebSearchResult,
   OpenAIResponsesLanguageModel,
   openaiResponsesSupportedUrls,
 } from './responses/openai-responses-language-model';
@@ -812,9 +812,7 @@ async function convertOpenAIBatchResult(
           type: 'tool-result',
           toolCallId: part.id,
           toolName: 'web_search',
-          ...(part.status === 'failed' || part.status === 'incomplete'
-            ? { isError: true, result: { status: part.status } }
-            : { result: mapWebSearchOutput(part.action) }),
+          ...mapWebSearchResult(part),
           dynamic: true,
         });
         break;

package/src/openai-language-model-capabilities.ts

@@ -24,7 +24,10 @@ export function getOpenAILanguageModelCapabilities(
   const isGptNanoModel = gptVersion?.variant?.startsWith('nano') ?? false;
   const isGpt6OrLaterModel = gptVersion != null && gptVersion.major >= 6;
 
-  const isGpt6SolOrLuna = modelId === 'gpt-6-sol' || modelId === 'gpt-6-luna';
+  const isGpt6OrLaterModelWithNoneEffort =
+    modelId === 'gpt-6-sol' ||
+    modelId === 'gpt-6-luna' ||
+    modelId === 'gpt-6.1-terra';
 
   const supportsFlexProcessing =
     (oSeriesVersion != null && oSeriesVersion >= 3) ||
@@ -45,10 +48,10 @@ export function getOpenAILanguageModelCapabilities(
     (gptVersion != null && gptVersion.major >= 5 && !isGptChatModel);
 
   // https://platform.openai.com/docs/guides/latest-model#gpt-5-1-parameter-compatibility
-  // GPT-5.1 through GPT-5.x and GPT-6 Sol/Luna support temperature,
-  // topP, and logProbs when reasoningEffort is none.
+  // GPT-5.1 through GPT-5.x, GPT-6 Sol/Luna, and GPT-6.1 Terra support
+  // temperature, topP, and logProbs when reasoningEffort is none.
   const supportsNonReasoningParameters =
-    isGpt6SolOrLuna ||
+    isGpt6OrLaterModelWithNoneEffort ||
     (!isGpt6OrLaterModel &&
       gptVersion != null &&
       (gptVersion.major > 5 ||
@@ -61,7 +64,7 @@ export function getOpenAILanguageModelCapabilities(
     supportsPriorityProcessing,
     supportsConfigurationUpdate: isGpt6OrLaterModel,
     supportsAsyncToolCalling: isGpt6OrLaterModel,
-    supportedReasoningEfforts: isGpt6SolOrLuna
+    supportedReasoningEfforts: isGpt6OrLaterModelWithNoneEffort
       ? ['none', 'low', 'medium', 'high', 'xhigh', 'max']
       : isGpt6OrLaterModel
         ? ['low', 'medium', 'high', 'xhigh', 'max']

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

@@ -59,6 +59,7 @@ export const openaiResponsesReasoningModelIds = [
   'gpt-5.6-terra',
   'gpt-6-astra',
   'gpt-6.1-sol',
+  'gpt-6.1-terra',
   'gpt-6-luna',
   'gpt-6-sol',
 ] as const;
@@ -135,6 +136,7 @@ export type OpenAIResponsesModelId =
   | 'gpt-5.6-terra'
   | 'gpt-6-astra'
   | 'gpt-6.1-sol'
+  | 'gpt-6.1-terra'
   | 'gpt-6-luna'
   | 'gpt-6-sol'
   | 'gpt-5-2025-08-07'

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

@@ -1275,9 +1275,7 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
             toolName: toolNameMapping.toCustomToolName(
               webSearchToolName ?? 'web_search',
             ),
-            ...(part.status === 'failed' || part.status === 'incomplete'
-              ? { isError: true, result: { status: part.status } }
-              : { result: mapWebSearchOutput(part.action) }),
+            ...mapWebSearchResult(part),
           });
 
           if (part.action?.type === 'search') {
@@ -2190,10 +2188,7 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
                   toolName: toolNameMapping.toCustomToolName(
                     webSearchToolName ?? 'web_search',
                   ),
-                  ...(value.item.status === 'failed' ||
-                  value.item.status === 'incomplete'
-                    ? { isError: true, result: { status: value.item.status } }
-                    : { result: mapWebSearchOutput(value.item.action) }),
+                  ...mapWebSearchResult(value.item),
                 });
 
                 if (value.item.action?.type === 'search') {
@@ -3180,7 +3175,21 @@ function isResponseOutputChunk(chunk: OpenAIResponsesChunk): boolean {
   );
 }
 
-export function mapWebSearchOutput(
+export function mapWebSearchResult({
+  action,
+  status,
+}: {
+  action?: OpenAIResponsesWebSearchAction | null;
+  status: string;
+}):
+  | { isError: true; result: { status: string } }
+  | { result: InferSchema<typeof webSearchOutputSchema> } {
+  return status === 'completed'
+    ? { result: mapWebSearchOutput(action) }
+    : { isError: true, result: { status } };
+}
+
+function mapWebSearchOutput(
   action: OpenAIResponsesWebSearchAction | null | undefined,
 ): InferSchema<typeof webSearchOutputSchema> {
   if (action == null) {

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

@@ -12,7 +12,9 @@ export type OpenAITranscriptionModelId =
   | 'gpt-4o-mini-transcribe-2025-12-15'
   | 'gpt-4o-transcribe'
   | 'gpt-4o-transcribe-diarize'
+  | 'gpt-live-transcribe'
   | 'gpt-realtime-whisper'
+  | 'gpt-transcribe'
   | (string & {});
 
 // https://platform.openai.com/docs/api-reference/audio/createTranscription
@@ -43,12 +45,9 @@ export const openAITranscriptionModelOptions = lazySchema(() =>
 
       /**
        * The timestamp granularities to populate for this transcription.
-       * @default ['segment']
+       * Defaults to ['segment'] when the response format is verbose_json.
        */
-      timestampGranularities: z
-        .array(z.enum(['word', 'segment']))
-        .default(['segment'])
-        .optional(),
+      timestampGranularities: z.array(z.enum(['word', 'segment'])).optional(),
 
       /**
        * The format of the transcription response.
@@ -73,7 +72,7 @@ export const openAITranscriptionModelOptions = lazySchema(() =>
         .optional(),
 
       /**
-       * Options for streaming transcription models such as `gpt-realtime-whisper`.
+       * Options for streaming transcription models such as `gpt-live-transcribe`.
        */
       streaming: z
         .object({

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

@@ -58,14 +58,43 @@ type OpenAIRealtimeTranscriptionEvent = {
 };
 
 /**
- * Realtime transcription model IDs stream over the realtime WebSocket
- * and do not support the REST transcription endpoint. Prefix matching
- * keeps dated snapshots (e.g. `gpt-realtime-whisper-2026-01-01`) working.
+ * Whether `modelId` is `base` or one of its dated snapshots
+ * (e.g. `gpt-realtime-whisper-2026-01-01`).
+ */
+function isModelOrSnapshot(modelId: string, base: string): boolean {
+  return modelId === base || modelId.startsWith(`${base}-`);
+}
+
+/**
+ * Model IDs that stream over the realtime transcription WebSocket and do not
+ * support the REST transcription endpoint.
  */
 function isRealtimeTranscriptionModelId(modelId: string): boolean {
   return (
-    modelId === 'gpt-realtime-whisper' ||
-    modelId.startsWith('gpt-realtime-whisper-')
+    isModelOrSnapshot(modelId, 'gpt-realtime-whisper') ||
+    isModelOrSnapshot(modelId, 'gpt-live-transcribe')
+  );
+}
+
+/**
+ * Model IDs that support streaming transcription. `gpt-transcribe` also
+ * supports the REST transcription endpoint.
+ */
+function supportsStreamingTranscription(modelId: string): boolean {
+  return (
+    isRealtimeTranscriptionModelId(modelId) ||
+    isModelOrSnapshot(modelId, 'gpt-transcribe')
+  );
+}
+
+/**
+ * Models whose realtime transcription session takes a `languages` list
+ * instead of the singular `language` field.
+ */
+function usesLanguagesList(modelId: string): boolean {
+  return (
+    isModelOrSnapshot(modelId, 'gpt-live-transcribe') ||
+    isModelOrSnapshot(modelId, 'gpt-transcribe')
   );
 }
 
@@ -197,54 +226,45 @@ export class OpenAITranscriptionModel implements TranscriptionModelV4 {
       `audio.${fileExtension}`,
     );
 
-    if (this.modelId === 'whisper-1') {
-      formData.append('response_format', 'verbose_json');
-    }
-
     const isDiarizationModel = this.modelId === 'gpt-4o-transcribe-diarize';
+
+    const responseFormat =
+      openAIOptions?.responseFormat ??
+      (isDiarizationModel
+        ? 'diarized_json'
+        : this.modelId === 'whisper-1'
+          ? 'verbose_json'
+          : undefined);
+
+    // Timestamps are only returned with verbose_json, so only default to
+    // segments when that format is used.
+    const timestampGranularities =
+      openAIOptions?.timestampGranularities ??
+      (responseFormat === 'verbose_json' ? ['segment'] : undefined);
+
     const chunkingStrategy =
       openAIOptions?.chunkingStrategy ??
       (isDiarizationModel ? 'auto' : undefined);
 
-    // Add provider-specific options
-    if (openAIOptions) {
-      const isGpt4oTranscribeModel = [
-        'gpt-4o-transcribe',
-        'gpt-4o-mini-transcribe',
-      ].includes(this.modelId);
-
-      const transcriptionModelOptions = {
-        include: openAIOptions.include,
-        language: openAIOptions.language,
-        prompt: openAIOptions.prompt,
-        // 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:
-            openAIOptions.responseFormat ??
-            (isDiarizationModel
-              ? 'diarized_json'
-              : isGpt4oTranscribeModel
-                ? 'json'
-                : 'verbose_json'),
-        }),
-        temperature: openAIOptions.temperature,
-        timestamp_granularities: openAIOptions.timestampGranularities,
-      };
+    const transcriptionModelOptions = {
+      response_format: responseFormat,
+      include: openAIOptions?.include,
+      language: openAIOptions?.language,
+      prompt: openAIOptions?.prompt,
+      temperature: openAIOptions?.temperature,
+      timestamp_granularities: timestampGranularities,
+    };
 
-      for (const [key, value] of Object.entries(transcriptionModelOptions)) {
-        if (value != null) {
-          if (Array.isArray(value)) {
-            for (const item of value) {
-              formData.append(`${key}[]`, String(item));
-            }
-          } else {
-            formData.append(key, String(value));
+    for (const [key, value] of Object.entries(transcriptionModelOptions)) {
+      if (value != null) {
+        if (Array.isArray(value)) {
+          for (const item of value) {
+            formData.append(`${key}[]`, String(item));
           }
+        } else {
+          formData.append(key, String(value));
         }
       }
-    } else if (isDiarizationModel) {
-      formData.append('response_format', 'diarized_json');
     }
 
     if (chunkingStrategy != null) {
@@ -344,7 +364,13 @@ export class OpenAITranscriptionModel implements TranscriptionModelV4 {
         })) ??
         [],
       language,
-      durationInSeconds: response.duration ?? undefined,
+      // Duration-billed json responses report the duration only in `usage`:
+      durationInSeconds:
+        response.duration ??
+        (response.usage?.type === 'duration' &&
+        typeof response.usage.seconds === 'number'
+          ? response.usage.seconds
+          : undefined),
       warnings,
       ...(response.usage != null && { usage: response.usage }),
       response: {
@@ -375,7 +401,7 @@ export class OpenAITranscriptionModel implements TranscriptionModelV4 {
   ): Promise<
     Awaited<ReturnType<NonNullable<TranscriptionModelV4['doStream']>>>
   > {
-    if (!isRealtimeTranscriptionModelId(this.modelId)) {
+    if (!supportsStreamingTranscription(this.modelId)) {
       throw new UnsupportedFunctionalityError({
         functionality: `streaming transcription with ${this.modelId}`,
       });
@@ -407,11 +433,22 @@ export class OpenAITranscriptionModel implements TranscriptionModelV4 {
       }
     }
 
+    // gpt-transcribe transcribes each committed turn, so it has no delay setting:
+    const supportsDelay = !isModelOrSnapshot(this.modelId, 'gpt-transcribe');
+    if (!supportsDelay && openAIOptions?.streaming?.delay != null) {
+      warnings.push({
+        type: 'unsupported',
+        feature: 'providerOptions.openai.streaming.delay',
+        details: `OpenAI streaming transcription with ${this.modelId} does not support delay.`,
+      });
+    }
+
     const headers = combineHeaders(this.config.headers?.(), options.headers);
     const sessionUpdate = buildOpenAIRealtimeTranscriptionSession({
       modelId: this.modelId,
       inputAudioFormat: options.inputAudioFormat,
       providerOptions: openAIOptions,
+      supportsDelay,
     });
 
     return {
@@ -600,10 +637,12 @@ function buildOpenAIRealtimeTranscriptionSession({
   modelId,
   inputAudioFormat,
   providerOptions,
+  supportsDelay,
 }: {
   modelId: string;
   inputAudioFormat: TranscriptionModelV4StreamOptions['inputAudioFormat'];
   providerOptions: OpenAITranscriptionModelOptions | undefined;
+  supportsDelay: boolean;
 }) {
   return {
     type: 'session.update',
@@ -620,9 +659,11 @@ function buildOpenAIRealtimeTranscriptionSession({
           transcription: {
             model: modelId,
             ...(providerOptions?.language != null
-              ? { language: providerOptions.language }
+              ? usesLanguagesList(modelId)
+                ? { languages: [providerOptions.language] }
+                : { language: providerOptions.language }
               : {}),
-            ...(providerOptions?.streaming?.delay != null
+            ...(supportsDelay && providerOptions?.streaming?.delay != null
               ? { delay: providerOptions.streaming.delay }
               : {}),
           },