@ai-sdk/google 4.0.25 → 4.0.26

package/docs/15-google.mdx

@@ -1133,6 +1133,57 @@ Gemini Live Translation accepts audio input and produces translated audio
 output. Text input, tools, and custom instructions are not supported by this
 model.
 
+## Translation Models
+
+<Note type="warning">Speech translation is an experimental feature.</Note>
+
+For server-side streaming translation, you can create models that call the
+[Gemini Live API](https://ai.google.dev/gemini-api/docs/live) using the
+`.translation()` factory method. Translation models are streaming-only and are
+used with
+[`experimental_streamTranslate`](/docs/reference/ai-sdk-core/stream-translate).
+
+The first argument is the model id e.g. `gemini-3.5-live-translate-preview`.
+
+```ts
+const model = google.translation('gemini-3.5-live-translate-preview');
+```
+
+```ts
+import { experimental_streamTranslate as streamTranslate } from 'ai';
+import { google } from '@ai-sdk/google';
+
+const result = streamTranslate({
+  model: google.translation('gemini-3.5-live-translate-preview'),
+  audio: audioStream, // ReadableStream<Uint8Array | string>
+  inputAudioFormat: { type: 'audio/pcm', rate: 16000 },
+  targetLanguage: 'es',
+});
+
+for await (const part of result.fullStream) {
+  if (part.type === 'output-text-delta') {
+    process.stdout.write(part.delta);
+  }
+}
+```
+
+The Gemini Live API auto-detects the source language (`sourceLanguage` is not
+supported) and always outputs 24kHz 16-bit PCM audio (`outputAudioFormat` is
+not supported). Unsupported settings surface as call warnings.
+
+The following provider options are available:
+
+- **echoTargetLanguage** _boolean_
+  Whether input audio already in the target language should be echoed instead
+  of producing silence.
+  Optional.
+
+### Model Capabilities
+
+| Model                               | Translated Audio | Translated Text | Source Transcript |
+| ----------------------------------- | ---------------- | --------------- | ----------------- |
+| `gemini-3.5-live-translate-preview` | <Check />        | <Check />       | <Check />         |
+
 ## Interactions API
 
 The [Gemini Interactions API](https://ai.google.dev/gemini-api/docs/interactions)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/google",
-  "version": "4.0.25",
+  "version": "4.0.26",
   "type": "module",
   "license": "Apache-2.0",
   "sideEffects": false,
@@ -36,7 +36,7 @@
   },
   "dependencies": {
     "@ai-sdk/provider": "4.0.4",
-    "@ai-sdk/provider-utils": "5.0.13"
+    "@ai-sdk/provider-utils": "5.0.14"
   },
   "devDependencies": {
     "@types/node": "22.19.19",

package/src/get-realtime-base-url.ts

@@ -0,0 +1,31 @@
+/**
+ * Returns the base URL for Live API (WebSocket) endpoints, stripping a
+ * trailing `/v1beta` or `/v1alpha` API version segment: Live API paths carry
+ * their own version and are not nested under it.
+ */
+export function getRealtimeBaseURL(baseURL: string): URL {
+  const url = new URL(baseURL);
+  const pathSegments = url.pathname.split('/');
+  const version = pathSegments.at(-1);
+
+  if (version === 'v1beta' || version === 'v1alpha') {
+    pathSegments.pop();
+    url.pathname = pathSegments.join('/') || '/';
+  }
+
+  return url;
+}
+
+/**
+ * Builds a `wss://` Live API WebSocket URL for the given bidi service path
+ * (e.g. `google.ai.generativelanguage.v1beta.GenerativeService.BidiGenerateContent`).
+ */
+export function getRealtimeWebSocketURL(
+  baseURL: string,
+  webSocketPath: string,
+): URL {
+  const url = getRealtimeBaseURL(baseURL);
+  url.protocol = url.protocol === 'https:' ? 'wss:' : 'ws:';
+  url.pathname = `${url.pathname.replace(/\/$/, '')}/ws/${webSocketPath}`;
+  return url;
+}

package/src/google-provider.ts

@@ -8,6 +8,7 @@ import type {
   Experimental_RealtimeFactoryV4 as RealtimeFactoryV4,
   Experimental_RealtimeFactoryV4GetTokenOptions as RealtimeFactoryV4GetTokenOptions,
   SpeechModelV4,
+  Experimental_SpeechTranslationModelV4 as SpeechTranslationModelV4,
 } from '@ai-sdk/provider';
 import {
   generateId,
@@ -15,6 +16,7 @@ import {
   withoutTrailingSlash,
   withUserAgentSuffix,
   type FetchFunction,
+  type WebSocketConstructor,
 } from '@ai-sdk/provider-utils';
 import { VERSION } from './version';
 import { GoogleEmbeddingModel } from './google-embedding-model';
@@ -40,6 +42,8 @@ import {
 import type { GoogleInteractionsModelId } from './interactions/google-interactions-language-model-options';
 import type { GoogleInteractionsAgentName } from './interactions/google-interactions-agent';
 import { GoogleRealtimeModel } from './realtime/google-realtime-model';
+import { GoogleTranslationModel } from './translation/google-translation-model';
+import type { GoogleTranslationModelId } from './translation/google-translation-model-options';
 
 export interface GoogleProvider extends ProviderV4 {
   (modelId: GoogleModelId): LanguageModelV4;
@@ -91,6 +95,18 @@ export interface GoogleProvider extends ProviderV4 {
    */
   videoModel(modelId: GoogleVideoModelId): Experimental_VideoModelV4;
 
+  /**
+   * Creates an experimental model for streaming speech translation.
+   */
+  translation(modelId: GoogleTranslationModelId): SpeechTranslationModelV4;
+
+  /**
+   * Creates an experimental model for streaming speech translation.
+   */
+  speechTranslationModel(
+    modelId: GoogleTranslationModelId,
+  ): SpeechTranslationModelV4;
+
   /**
    * Creates a model for speech generation (text-to-speech).
    */
@@ -152,6 +168,12 @@ export interface GoogleProviderSettings {
    */
   generateId?: () => string;
 
+  /**
+   * Custom WebSocket implementation. This is useful for testing or for
+   * runtimes that need a WebSocket constructor with header support.
+   */
+  webSocket?: WebSocketConstructor;
+
   /**
    * Custom provider name
    * Defaults to 'google.generative-ai'.
@@ -288,6 +310,14 @@ export function createGoogle(
       fetch: options.fetch,
     });
 
+  const createTranslationModel = (modelId: GoogleTranslationModelId) =>
+    new GoogleTranslationModel(modelId, {
+      provider: `${providerName}.translation`,
+      baseURL,
+      headers: getHeaders,
+      webSocket: options.webSocket,
+    });
+
   const createSpeechModel = (modelId: GoogleSpeechModelId) =>
     new GoogleSpeechModel(modelId, {
       provider: `${providerName}.speech`,
@@ -358,6 +388,8 @@ export function createGoogle(
   provider.files = createFiles;
   provider.speech = createSpeechModel;
   provider.speechModel = createSpeechModel;
+  provider.translation = createTranslationModel;
+  provider.speechTranslationModel = createTranslationModel;
   provider.interactions = createInteractionsModel;
   provider.tools = googleTools;
 

package/src/index.ts

@@ -60,5 +60,11 @@ export type {
   GoogleRealtimeModelId as Experimental_GoogleRealtimeModelId,
   GoogleRealtimeModelOptions as Experimental_GoogleRealtimeModelOptions,
 } from './realtime/google-realtime-model-options';
+export { GoogleTranslationModel as Experimental_GoogleTranslationModel } from './translation/google-translation-model';
+export type { GoogleTranslationModelConfig as Experimental_GoogleTranslationModelConfig } from './translation/google-translation-model';
+export type {
+  GoogleTranslationModelId as Experimental_GoogleTranslationModelId,
+  GoogleTranslationModelOptions as Experimental_GoogleTranslationModelOptions,
+} from './translation/google-translation-model-options';
 
 export { VERSION } from './version';

package/src/realtime/google-realtime-model.ts

@@ -7,6 +7,10 @@ import type {
   Experimental_RealtimeModelV4SessionConfig as RealtimeModelV4SessionConfig,
 } from '@ai-sdk/provider';
 import type { FetchFunction } from '@ai-sdk/provider-utils';
+import {
+  getRealtimeBaseURL,
+  getRealtimeWebSocketURL,
+} from '../get-realtime-base-url';
 import {
   GoogleRealtimeEventMapper,
   buildGoogleSessionConfig,
@@ -15,19 +19,6 @@ import {
 const realtimeWebSocketPath =
   'google.ai.generativelanguage.v1alpha.GenerativeService.BidiGenerateContentConstrained';
 
-function getRealtimeBaseURL(baseURL: string): URL {
-  const url = new URL(baseURL);
-  const pathSegments = url.pathname.split('/');
-  const version = pathSegments.at(-1);
-
-  if (version === 'v1beta' || version === 'v1alpha') {
-    pathSegments.pop();
-    url.pathname = pathSegments.join('/') || '/';
-  }
-
-  return url;
-}
-
 function getAuthTokensURL(baseURL: string): string {
   const url = getRealtimeBaseURL(baseURL);
   url.pathname = `${url.pathname.replace(/\/$/, '')}/v1alpha/auth_tokens`;
@@ -35,10 +26,7 @@ function getAuthTokensURL(baseURL: string): string {
 }
 
 function getWebSocketURL(baseURL: string): string {
-  const url = getRealtimeBaseURL(baseURL);
-  url.protocol = url.protocol === 'https:' ? 'wss:' : 'ws:';
-  url.pathname = `${url.pathname.replace(/\/$/, '')}/ws/${realtimeWebSocketPath}`;
-  return url.toString();
+  return getRealtimeWebSocketURL(baseURL, realtimeWebSocketPath).toString();
 }
 
 export type GoogleRealtimeModelConfig = {

package/src/translation/google-translation-model-options.ts

@@ -0,0 +1,26 @@
+import {
+  lazySchema,
+  zodSchema,
+  type InferSchema,
+} from '@ai-sdk/provider-utils';
+import { z } from 'zod/v4';
+
+export type GoogleTranslationModelId =
+  | 'gemini-3.5-live-translate-preview'
+  | (string & {});
+
+export const googleTranslationModelOptions = lazySchema(() =>
+  zodSchema(
+    z.object({
+      /**
+       * Whether input audio already in the target language should be echoed
+       * instead of producing silence.
+       */
+      echoTargetLanguage: z.boolean().optional(),
+    }),
+  ),
+);
+
+export type GoogleTranslationModelOptions = InferSchema<
+  typeof googleTranslationModelOptions
+>;