@ai-sdk/google 4.0.0-canary.80 → 4.0.0

package/docs/15-google.mdx

@@ -1091,6 +1091,45 @@ Google realtime models may require provider-specific audio formats, depending
 on the model and modality. See [Realtime](/docs/ai-sdk-core/realtime) for the
 complete setup and tool calling pattern.
 
+### Live Translation
+
+Use `gemini-3.5-live-translate-preview` for low-latency speech-to-speech
+translation. Configure the target language with
+`providerOptions.google.translationConfig` in the realtime session config you
+use when creating the token:
+
+```ts
+import {
+  google,
+  type Experimental_GoogleRealtimeModelOptions as GoogleRealtimeModelOptions,
+} from '@ai-sdk/google';
+
+const token = await google.experimental_realtime.getToken({
+  model: 'gemini-3.5-live-translate-preview',
+  sessionConfig: {
+    outputModalities: ['audio'],
+    inputAudioTranscription: {},
+    outputAudioTranscription: {},
+    providerOptions: {
+      google: {
+        translationConfig: {
+          targetLanguageCode: 'pl',
+          echoTargetLanguage: true,
+        },
+      } satisfies GoogleRealtimeModelOptions,
+    },
+  },
+});
+```
+
+Set `targetLanguageCode` to a BCP-47 language code (defaults to `en`). When
+`echoTargetLanguage` is `true`, input audio already in the target language is
+echoed instead of producing silence.
+
+Gemini Live Translation accepts audio input and produces translated audio
+output. Text input, tools, and custom instructions are not supported by this
+model.
+
 ## Interactions API
 
 The [Gemini Interactions API](https://ai.google.dev/gemini-api/docs/interactions)
@@ -1977,7 +2016,7 @@ The `voice` argument can be set to one of Gemini's [30 prebuilt voices](https://
 e.g. `Kore`, `Puck`, `Zephyr`, or `Charon`. Voice names are case-sensitive. It defaults to `Kore`.
 
 ```ts highlight="6"
-import { experimental_generateSpeech as generateSpeech } from 'ai';
+import { generateSpeech } from 'ai';
 import { google } from '@ai-sdk/google';
 
 const result = await generateSpeech({
@@ -2000,7 +2039,7 @@ For multi-speaker dialogue, pass a `multiSpeakerVoiceConfig` through `providerOp
 name must match a name used in the input text. When set, it overrides the top-level `voice`.
 
 ```ts highlight="8-23"
-import { experimental_generateSpeech as generateSpeech } from 'ai';
+import { generateSpeech } from 'ai';
 import { google, type GoogleSpeechModelOptions } from '@ai-sdk/google';
 
 const result = await generateSpeech({

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/google",
-  "version": "4.0.0-canary.80",
+  "version": "4.0.0",
   "type": "module",
   "license": "Apache-2.0",
   "sideEffects": false,
@@ -35,15 +35,15 @@
     }
   },
   "dependencies": {
-    "@ai-sdk/provider": "4.0.0-canary.18",
-    "@ai-sdk/provider-utils": "5.0.0-canary.47"
+    "@ai-sdk/provider": "4.0.0",
+    "@ai-sdk/provider-utils": "5.0.0"
   },
   "devDependencies": {
     "@types/node": "22.19.19",
     "tsup": "^8.5.1",
     "typescript": "5.8.3",
     "zod": "3.25.76",
-    "@ai-sdk/test-server": "2.0.0-canary.6",
+    "@ai-sdk/test-server": "2.0.0",
     "@vercel/ai-tsconfig": "0.0.0"
   },
   "peerDependencies": {

package/src/google-json-accumulator.ts

@@ -276,6 +276,35 @@ function parsePath(rawPath: string): Array<string | number> {
   return segments;
 }
 
+const hasOwn = Object.prototype.hasOwnProperty;
+
+/**
+ * Checks only direct properties so path traversal never follows the prototype chain.
+ */
+function hasOwnProperty(
+  obj: Record<string | number, unknown>,
+  key: string | number,
+): boolean {
+  return hasOwn.call(obj, key);
+}
+
+/**
+ * Defines path values as own data properties so special keys like `__proto__`
+ * cannot invoke prototype setters while accumulating streamed arguments.
+ */
+function defineOwnProperty(
+  obj: Record<string | number, unknown>,
+  key: string | number,
+  value: unknown,
+): void {
+  Object.defineProperty(obj, key, {
+    value,
+    enumerable: true,
+    configurable: true,
+    writable: true,
+  });
+}
+
 /**
  * Traverses a nested object along the given path segments and returns the leaf value.
  *
@@ -289,7 +318,9 @@ function getNestedValue(
   let current: unknown = obj;
   for (const pathSegment of segments) {
     if (current == null || typeof current !== 'object') return undefined;
-    current = (current as Record<string | number, unknown>)[pathSegment];
+    const currentRecord = current as Record<string | number, unknown>;
+    if (!hasOwnProperty(currentRecord, pathSegment)) return undefined;
+    current = currentRecord[pathSegment];
   }
   return current;
 }
@@ -309,12 +340,16 @@ function setNestedValue(
   for (let i = 0; i < segments.length - 1; i++) {
     const pathSegment = segments[i];
     const nextSeg = segments[i + 1];
-    if (current[pathSegment] == null) {
-      current[pathSegment] = typeof nextSeg === 'number' ? [] : {};
+    if (!hasOwnProperty(current, pathSegment) || current[pathSegment] == null) {
+      defineOwnProperty(
+        current,
+        pathSegment,
+        typeof nextSeg === 'number' ? [] : {},
+      );
     }
     current = current[pathSegment] as Record<string | number, unknown>;
   }
-  current[segments[segments.length - 1]] = value;
+  defineOwnProperty(current, segments[segments.length - 1], value);
 }
 
 /**

package/src/google-video-model.ts

@@ -9,6 +9,7 @@ import {
   createJsonResponseHandler,
   delay,
   getFromApi,
+  isSameOrigin,
   parseProviderOptions,
   postJsonToApi,
   resolve,
@@ -263,10 +264,13 @@ export class GoogleVideoModel implements Experimental_VideoModelV4 {
     for (const generatedSample of response.generateVideoResponse
       .generatedSamples) {
       if (generatedSample.video?.uri) {
-        // Append API key to URL for authentication during download
-        const urlWithAuth = apiKey
-          ? `${generatedSample.video.uri}${generatedSample.video.uri.includes('?') ? '&' : '?'}key=${apiKey}`
-          : generatedSample.video.uri;
+        // Append the API key to the download URL for authentication, but only
+        // when the response-supplied URI stays on the provider's own origin —
+        // otherwise the key would leak to whatever host the response names.
+        const urlWithAuth =
+          apiKey && isSameOrigin(generatedSample.video.uri, this.config.baseURL)
+            ? `${generatedSample.video.uri}${generatedSample.video.uri.includes('?') ? '&' : '?'}key=${apiKey}`
+            : generatedSample.video.uri;
 
         videos.push({
           type: 'url',

package/src/index.ts

@@ -56,5 +56,9 @@ export type {
 } from './google-provider';
 export { GoogleRealtimeModel as Experimental_GoogleRealtimeModel } from './realtime/google-realtime-model';
 export type { GoogleRealtimeModelConfig as Experimental_GoogleRealtimeModelConfig } from './realtime/google-realtime-model';
+export type {
+  GoogleRealtimeModelId as Experimental_GoogleRealtimeModelId,
+  GoogleRealtimeModelOptions as Experimental_GoogleRealtimeModelOptions,
+} from './realtime/google-realtime-model-options';
 
 export { VERSION } from './version';

package/src/interactions/google-interactions-language-model.ts

@@ -459,7 +459,6 @@ export class GoogleInteractionsLanguageModel implements LanguageModelV4 {
     const url = `${this.config.baseURL}/interactions`;
 
     const mergedHeaders = combineHeaders(
-      INTERACTIONS_API_REVISION_HEADER,
       this.config.headers ? await resolve(this.config.headers) : undefined,
       options.headers,
     );
@@ -588,7 +587,6 @@ export class GoogleInteractionsLanguageModel implements LanguageModelV4 {
     const url = `${this.config.baseURL}/interactions`;
 
     const mergedHeaders = combineHeaders(
-      INTERACTIONS_API_REVISION_HEADER,
       this.config.headers ? await resolve(this.config.headers) : undefined,
       options.headers,
     );
@@ -757,15 +755,6 @@ export class GoogleInteractionsLanguageModel implements LanguageModelV4 {
   }
 }
 
-/*
- * Pins the Interactions API revision the SDK targets. Sent on every request
- * the model issues so model-id calls, agent calls, polling, SSE reconnects,
- * and cancellation all hit the same schema.
- */
-const INTERACTIONS_API_REVISION_HEADER: Record<string, string> = {
-  'Api-Revision': '2026-05-20',
-};
-
 function pruneUndefined<T extends Record<string, unknown>>(obj: T): T {
   const result: Record<string, unknown> = {};
   for (const [key, value] of Object.entries(obj)) {

package/src/realtime/google-realtime-event-mapper.ts

@@ -8,6 +8,7 @@ import type {
 import { safeParseJSON } from '@ai-sdk/provider-utils';
 import { convertJSONSchemaToOpenAPISchema } from '../convert-json-schema-to-openapi-schema';
 import { getModelPath } from '../get-model-path';
+import type { GoogleRealtimeModelOptions } from './google-realtime-model-options';
 
 type GoogleRealtimeFunctionCall = {
   id: string;
@@ -38,6 +39,10 @@ type GoogleRealtimeWireEvent = {
   inputTranscription?: { text?: string };
 };
 
+function isRecord(value: unknown): value is Record<string, unknown> {
+  return value != null && typeof value === 'object' && !Array.isArray(value);
+}
+
 /**
  * Stateful event mapper for Google's Gemini Live API.
  *
@@ -268,6 +273,12 @@ export class GoogleRealtimeEventMapper {
         };
 
       case 'input-audio-commit':
+        return {
+          realtimeInput: {
+            audioStreamEnd: true,
+          },
+        };
+
       case 'input-audio-clear':
       case 'response-create':
       case 'response-cancel':
@@ -375,8 +386,25 @@ export function buildGoogleSessionConfig(
     setup.outputAudioTranscription = {};
   }
 
-  if (config?.providerOptions != null) {
-    Object.assign(setup, config.providerOptions);
+  if (config?.providerOptions == null) {
+    return setup;
+  }
+
+  const { google, ...providerOptions } = config.providerOptions;
+  Object.assign(setup, providerOptions);
+
+  const googleOptions = isRecord(google)
+    ? (google as GoogleRealtimeModelOptions)
+    : undefined;
+
+  if (googleOptions?.translationConfig != null) {
+    const target = isRecord(setup.generationConfig)
+      ? setup.generationConfig
+      : generationConfig;
+    setup.generationConfig = {
+      ...target,
+      translationConfig: googleOptions.translationConfig,
+    };
   }
 
   return setup;

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

@@ -1,3 +1,23 @@
 export type GoogleRealtimeModelId = string;
 
-export type GoogleRealtimeModelOptions = Record<string, never>;
+export type GoogleRealtimeModelOptions = {
+  /**
+   * Gemini Live Translation configuration.
+   *
+   * Required for `gemini-3.5-live-translate-preview` when translating speech
+   * to a target language.
+   */
+  translationConfig?: {
+    /**
+     * BCP-47 language code of the language to translate into.
+     * Defaults to `en` in the Gemini API.
+     */
+    targetLanguageCode?: string;
+
+    /**
+     * Whether input audio already in the target language should be echoed
+     * instead of producing silence.
+     */
+    echoTargetLanguage?: boolean;
+  };
+};

package/src/realtime/index.ts

@@ -1,2 +1,6 @@
 export { GoogleRealtimeModel as Experimental_GoogleRealtimeModel } from './google-realtime-model';
 export type { GoogleRealtimeModelConfig as Experimental_GoogleRealtimeModelConfig } from './google-realtime-model';
+export type {
+  GoogleRealtimeModelId as Experimental_GoogleRealtimeModelId,
+  GoogleRealtimeModelOptions as Experimental_GoogleRealtimeModelOptions,
+} from './google-realtime-model-options';