@ai-sdk/google 3.0.74 → 3.0.77

package/docs/15-google-generative-ai.mdx

@@ -1066,6 +1066,7 @@ The following Zod features are known to not work with Google Generative AI:
 
 | Model                                 | Image Input         | Object Generation   | Tool Usage          | Tool Streaming      | Google Search       | URL Context         |
 | ------------------------------------- | ------------------- | ------------------- | ------------------- | ------------------- | ------------------- | ------------------- |
+| `gemini-3.5-flash`                    | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
 | `gemini-3.1-pro-preview`              | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
 | `gemini-3.1-flash-image-preview`      | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
 | `gemini-3.1-flash-lite-preview`       | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
@@ -1179,21 +1180,28 @@ The following optional provider options are available:
   Whether the model returns synthesized thought summaries on reasoning
   parts. Defaults to the API default.
 
-- **imageConfig** _\{ aspectRatio?: string; imageSize?: '1K' | '2K' | '4K' | '512' \}_
+- **responseFormat** _Array\<\{ type: 'text' | 'image' | 'audio'; mimeType?: string; schema?: unknown; aspectRatio?: string; imageSize?: '1K' \| '2K' \| '4K' \| '512' \}\>_
 
-  Image generation configuration when `responseModalities` includes
-  `'image'`. `aspectRatio` accepts `1:1`, `2:3`, `3:2`, `3:4`, `4:3`,
-  `4:5`, `5:4`, `9:16`, `16:9`, `21:9`, `1:8`, `8:1`, `1:4`, `4:1`.
+  Output-format entries that map directly to the API's `response_format`
+  array. Use this for fine-grained control over image, audio, or non-JSON
+  text outputs (e.g. `aspectRatio` and `imageSize` for image generation).
+  The AI SDK call-level `responseFormat: { type: 'json', schema }` still
+  drives JSON-mode automatically and prepends a matching text entry;
+  entries listed here are appended.
 
-- **mediaResolution** _'low' | 'medium' | 'high' | 'ultra_high'_
+  `aspectRatio` accepts `1:1`, `2:3`, `3:2`, `3:4`, `4:3`, `4:5`, `5:4`,
+  `9:16`, `16:9`, `21:9`, `1:8`, `8:1`, `1:4`, `4:1`.
 
-  Media resolution applied to image inputs / outputs.
+- **imageConfig** _\{ aspectRatio?: string; imageSize?: '1K' | '2K' | '4K' | '512' \}_ (deprecated)
 
-- **responseModalities** _Array\<'text' | 'image' | 'audio' | 'video' | 'document'\>_
+  Use **responseFormat** with a `{ type: 'image', ... }` entry instead.
+  Retained for backwards compatibility; the SDK translates `imageConfig`
+  into a matching `response_format` image entry and emits a warning when
+  set. Ignored when `responseFormat` already supplies an image entry.
 
-  The modalities the model may emit. Defaults to text-only. Pass
-  `['image']` (or `['text', 'image']`) to enable native image output. See
-  [Image output](#image-output-via-interactions).
+- **mediaResolution** _'low' | 'medium' | 'high' | 'ultra_high'_
+
+  Media resolution applied to image inputs / outputs.
 
 - **serviceTier** _'flex' | 'standard' | 'priority'_
 
@@ -1346,9 +1354,10 @@ const { text, toolCalls } = await generateText({
 
 ### Image output via Interactions
 
-Set `responseModalities: ['image']` on a Gemini image-capable model to get
-images as `LanguageModelV4FilePart` files in the response. No tool wrapping
-is required.
+Add a `{ type: 'image' }` entry to `responseFormat` on a Gemini
+image-capable model to get images as `LanguageModelV4FilePart` files in
+the response. No tool wrapping is required, and the entry doubles as the
+place to set `aspectRatio`, `imageSize`, and `mimeType`.
 
 ```ts
 import { google } from '@ai-sdk/google';
@@ -1359,7 +1368,7 @@ const result = await generateText({
   prompt: 'Generate an image of a comic cat in a spaceship.',
   providerOptions: {
     google: {
-      responseModalities: ['image'],
+      responseFormat: [{ type: 'image' }],
     },
   },
 });
@@ -1371,6 +1380,54 @@ for (const file of result.files) {
 }
 ```
 
+To control aspect ratio, image size, or output mime type, add those
+fields to the same image entry:
+
+```ts
+const result = await generateText({
+  model: google.interactions('gemini-3-pro-image-preview'),
+  prompt: 'Generate a high-quality landscape photo of mountains at sunset.',
+  providerOptions: {
+    google: {
+      responseFormat: [
+        {
+          type: 'image',
+          aspectRatio: '16:9',
+          imageSize: '4K',
+        },
+      ],
+    },
+  },
+});
+```
+
+For multimodal output, list one entry per modality. The model returns
+text in `result.text` and the accompanying image(s) in `result.files`:
+
+```ts
+import { google } from '@ai-sdk/google';
+import { generateText } from 'ai';
+
+const result = await generateText({
+  model: google.interactions('gemini-2.5-flash-image'),
+  prompt:
+    'Tell me a three sentence bedtime story about a unicorn, accompanied by a suitable illustration.',
+  providerOptions: {
+    google: {
+      responseFormat: [
+        { type: 'text' },
+        { type: 'image', aspectRatio: '16:9' },
+      ],
+    },
+  },
+});
+
+console.log(result.text);
+
+const images = result.files.filter(file => file.mediaType.startsWith('image/'));
+// images[0].uint8Array | images[0].base64 | images[0].mediaType
+```
+
 Iterative image editing pairs naturally with stateful chaining — keep
 `previousInteractionId` set across turns and the model edits its prior
 output:
@@ -1384,7 +1441,7 @@ const model = google.interactions('gemini-3-pro-image-preview');
 const turn1 = await generateText({
   model,
   prompt: 'Generate an image of a comic cat in a spaceship.',
-  providerOptions: { google: { responseModalities: ['image'] } },
+  providerOptions: { google: { responseFormat: [{ type: 'image' }] } },
 });
 
 const interactionId = turn1.providerMetadata?.google?.interactionId as
@@ -1396,7 +1453,7 @@ const turn2 = await generateText({
   prompt: 'now make the cat red',
   providerOptions: {
     google: {
-      responseModalities: ['image'],
+      responseFormat: [{ type: 'image' }],
       previousInteractionId: interactionId,
     },
   },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/google",
-  "version": "3.0.74",
+  "version": "3.0.77",
   "license": "Apache-2.0",
   "sideEffects": false,
   "main": "./dist/index.js",

package/src/google-generative-ai-language-model.ts

@@ -37,7 +37,6 @@ import { getModelPath } from './get-model-path';
 import { googleFailedResponseHandler } from './google-error';
 import {
   googleLanguageModelOptions,
-  VertexServiceTierMap,
   type GoogleGenerativeAIModelId,
 } from './google-generative-ai-options';
 import type {
@@ -155,11 +154,43 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
       });
     }
 
-    // Vertex API requires another service tier format.
-    let sanitizedServiceTier: string | undefined = googleOptions?.serviceTier;
     if (googleOptions?.serviceTier && isVertexProvider) {
-      sanitizedServiceTier = VertexServiceTierMap[googleOptions.serviceTier];
+      warnings.push({
+        type: 'other',
+        message:
+          "'serviceTier' is a Gemini API option and is not supported on Vertex AI. " +
+          "Use 'sharedRequestType' (and optionally 'requestType') instead. See " +
+          'https://docs.cloud.google.com/vertex-ai/generative-ai/docs/priority-paygo',
+      });
     }
+    if (
+      (googleOptions?.sharedRequestType || googleOptions?.requestType) &&
+      !isVertexProvider
+    ) {
+      warnings.push({
+        type: 'other',
+        message:
+          "'sharedRequestType' and 'requestType' are Vertex AI options and " +
+          `are ignored with the current Google provider (${this.config.provider}).`,
+      });
+    }
+
+    const vertexPaygoHeaders: Record<string, string> | undefined =
+      isVertexProvider &&
+      (googleOptions?.sharedRequestType || googleOptions?.requestType)
+        ? {
+            ...(googleOptions.sharedRequestType && {
+              'X-Vertex-AI-LLM-Shared-Request-Type':
+                googleOptions.sharedRequestType,
+            }),
+            ...(googleOptions.requestType && {
+              'X-Vertex-AI-LLM-Request-Type': googleOptions.requestType,
+            }),
+          }
+        : undefined;
+    const bodyServiceTier = isVertexProvider
+      ? undefined
+      : googleOptions?.serviceTier;
 
     const isGemmaModel = this.modelId.toLowerCase().startsWith('gemma-');
     const supportsFunctionResponseParts = this.modelId.startsWith('gemini-3');
@@ -253,21 +284,24 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
         toolConfig,
         cachedContent: googleOptions?.cachedContent,
         labels: googleOptions?.labels,
-        serviceTier: sanitizedServiceTier,
+        serviceTier: bodyServiceTier,
       },
       warnings: [...warnings, ...toolWarnings],
       providerOptionsName,
+      extraHeaders: vertexPaygoHeaders,
     };
   }
 
   async doGenerate(
     options: LanguageModelV3CallOptions,
   ): Promise<LanguageModelV3GenerateResult> {
-    const { args, warnings, providerOptionsName } = await this.getArgs(options);
+    const { args, warnings, providerOptionsName, extraHeaders } =
+      await this.getArgs(options);
 
     const mergedHeaders = combineHeaders(
       await resolve(this.config.headers),
       options.headers,
+      extraHeaders,
     );
 
     const {
@@ -464,7 +498,7 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
           safetyRatings: candidate.safetyRatings ?? null,
           usageMetadata: usageMetadata ?? null,
           finishMessage: candidate.finishMessage ?? null,
-          serviceTier: response.serviceTier ?? null,
+          serviceTier: responseHeaders?.['x-gemini-service-tier'] ?? null,
         } satisfies GoogleGenerativeAIProviderMetadata,
       },
       request: { body: args },
@@ -479,14 +513,13 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
   async doStream(
     options: LanguageModelV3CallOptions,
   ): Promise<LanguageModelV3StreamResult> {
-    const { args, warnings, providerOptionsName } = await this.getArgs(
-      options,
-      { isStreaming: true },
-    );
+    const { args, warnings, providerOptionsName, extraHeaders } =
+      await this.getArgs(options, { isStreaming: true });
 
     const headers = combineHeaders(
       await resolve(this.config.headers),
       options.headers,
+      extraHeaders,
     );
 
     const { responseHeaders, value: response } = await postJsonToApi({
@@ -509,7 +542,8 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
     let providerMetadata: SharedV3ProviderMetadata | undefined = undefined;
     let lastGroundingMetadata: GroundingMetadataSchema | null = null;
     let lastUrlContextMetadata: UrlContextMetadataSchema | null = null;
-    let serviceTier: string | null = null;
+    const serviceTier: string | null =
+      responseHeaders?.['x-gemini-service-tier'] ?? null;
 
     const generateId = this.config.generateId;
     let hasToolCalls = false;
@@ -533,6 +567,42 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
       providerMetadata?: SharedV3ProviderMetadata;
     }> = [];
 
+    const finishActiveStreamingToolCall = (
+      controller: TransformStreamDefaultController<LanguageModelV3StreamPart>,
+    ) => {
+      const active = activeStreamingToolCalls.pop();
+      if (active == null) {
+        return;
+      }
+
+      const { finalJSON, closingDelta } = active.accumulator.finalize();
+
+      if (closingDelta.length > 0) {
+        controller.enqueue({
+          type: 'tool-input-delta',
+          id: active.toolCallId,
+          delta: closingDelta,
+          providerMetadata: active.providerMetadata,
+        });
+      }
+
+      controller.enqueue({
+        type: 'tool-input-end',
+        id: active.toolCallId,
+        providerMetadata: active.providerMetadata,
+      });
+
+      controller.enqueue({
+        type: 'tool-call',
+        toolCallId: active.toolCallId,
+        toolName: active.toolName,
+        input: finalJSON,
+        providerMetadata: active.providerMetadata,
+      });
+
+      hasToolCalls = true;
+    };
+
     return {
       stream: response.pipeThrough(
         new TransformStream<
@@ -561,10 +631,6 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
               usage = usageMetadata;
             }
 
-            if (value.serviceTier != null) {
-              serviceTier = value.serviceTier;
-            }
-
             const candidate = value.candidates?.[0];
 
             // sometimes the API returns an empty candidates array
@@ -824,10 +890,7 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
                   part.functionCall.willContinue !== true;
 
                 if (isStreamingChunk) {
-                  if (
-                    part.functionCall.name != null &&
-                    part.functionCall.willContinue === true
-                  ) {
+                  if (part.functionCall.name != null) {
                     const toolCallId = part.functionCall.id ?? generateId();
                     const accumulator = new GoogleJSONAccumulator();
                     activeStreamingToolCalls.push({
@@ -845,9 +908,10 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
                     });
 
                     if (part.functionCall.partialArgs != null) {
-                      const { textDelta } = accumulator.processPartialArgs(
-                        part.functionCall.partialArgs as PartialArg[],
-                      );
+                      const partialArgs = part.functionCall
+                        .partialArgs as PartialArg[];
+                      const { textDelta } =
+                        accumulator.processPartialArgs(partialArgs);
                       if (textDelta.length > 0) {
                         controller.enqueue({
                           type: 'tool-input-delta',
@@ -856,6 +920,12 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
                           providerMetadata: providerMeta,
                         });
                       }
+                      if (
+                        part.functionCall.willContinue !== true &&
+                        partialArgs.every(arg => arg.willContinue !== true)
+                      ) {
+                        finishActiveStreamingToolCall(controller);
+                      }
                     }
                   } else if (
                     part.functionCall.partialArgs != null &&
@@ -865,9 +935,10 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
                       activeStreamingToolCalls[
                         activeStreamingToolCalls.length - 1
                       ];
-                    const { textDelta } = active.accumulator.processPartialArgs(
-                      part.functionCall.partialArgs as PartialArg[],
-                    );
+                    const partialArgs = part.functionCall
+                      .partialArgs as PartialArg[];
+                    const { textDelta } =
+                      active.accumulator.processPartialArgs(partialArgs);
                     if (textDelta.length > 0) {
                       controller.enqueue({
                         type: 'tool-input-delta',
@@ -876,39 +947,18 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
                         providerMetadata: providerMeta,
                       });
                     }
+                    if (
+                      part.functionCall.willContinue !== true &&
+                      partialArgs.every(arg => arg.willContinue !== true)
+                    ) {
+                      finishActiveStreamingToolCall(controller);
+                    }
                   }
                 } else if (
                   isTerminalChunk &&
                   activeStreamingToolCalls.length > 0
                 ) {
-                  const active = activeStreamingToolCalls.pop()!;
-                  const { finalJSON, closingDelta } =
-                    active.accumulator.finalize();
-
-                  if (closingDelta.length > 0) {
-                    controller.enqueue({
-                      type: 'tool-input-delta',
-                      id: active.toolCallId,
-                      delta: closingDelta,
-                      providerMetadata: active.providerMetadata,
-                    });
-                  }
-
-                  controller.enqueue({
-                    type: 'tool-input-end',
-                    id: active.toolCallId,
-                    providerMetadata: active.providerMetadata,
-                  });
-
-                  controller.enqueue({
-                    type: 'tool-call',
-                    toolCallId: active.toolCallId,
-                    toolName: active.toolName,
-                    input: finalJSON,
-                    providerMetadata: active.providerMetadata,
-                  });
-
-                  hasToolCalls = true;
+                  finishActiveStreamingToolCall(controller);
                 } else if (isCompleteCall) {
                   const toolCallId = part.functionCall.id ?? generateId();
                   const toolName = part.functionCall.name!;
@@ -1377,7 +1427,6 @@ const responseSchema = lazySchema(() =>
           safetyRatings: z.array(getSafetyRatingSchema()).nullish(),
         })
         .nullish(),
-      serviceTier: z.string().nullish(),
     }),
   ),
 );
@@ -1433,7 +1482,6 @@ const chunkSchema = lazySchema(() =>
           safetyRatings: z.array(getSafetyRatingSchema()).nullish(),
         })
         .nullish(),
-      serviceTier: z.string().nullish(),
     }),
   ),
 );

package/src/google-generative-ai-options.ts

@@ -30,6 +30,7 @@ export type GoogleGenerativeAIModelId =
   | 'gemini-3.1-flash-image-preview'
   | 'gemini-3.1-flash-lite-preview'
   | 'gemini-3.1-flash-tts-preview'
+  | 'gemini-3.5-flash'
   // latest version
   // https://ai.google.dev/gemini-api/docs/models#latest
   | 'gemini-pro-latest'
@@ -206,9 +207,31 @@ export const googleLanguageModelOptions = lazySchema(() =>
       streamFunctionCallArguments: z.boolean().optional(),
 
       /**
-       * Optional. The service tier to use for the request.
+       * Optional. The service tier to use for the request. Sent as the
+       * `serviceTier` body field. Gemini API only.
        */
       serviceTier: z.enum(['standard', 'flex', 'priority']).optional(),
+
+      /**
+       * Optional. Vertex AI only. Sent as the
+       * `X-Vertex-AI-LLM-Shared-Request-Type` request header to select a
+       * shared (PayGo) tier. With Provisioned Throughput allocated and
+       * `requestType` unset, the request falls back to this tier only if
+       * PT capacity is exhausted.
+       *
+       * https://docs.cloud.google.com/vertex-ai/generative-ai/docs/priority-paygo
+       * https://docs.cloud.google.com/vertex-ai/generative-ai/docs/flex-paygo
+       */
+      sharedRequestType: z.enum(['priority', 'flex', 'standard']).optional(),
+
+      /**
+       * Optional. Vertex AI only. Sent as the `X-Vertex-AI-LLM-Request-Type`
+       * request header. Set to `'shared'` together with `sharedRequestType`
+       * to bypass Provisioned Throughput entirely.
+       *
+       * https://docs.cloud.google.com/vertex-ai/generative-ai/docs/priority-paygo
+       */
+      requestType: z.enum(['shared']).optional(),
     }),
   ),
 );
@@ -216,10 +239,3 @@ export const googleLanguageModelOptions = lazySchema(() =>
 export type GoogleLanguageModelOptions = InferSchema<
   typeof googleLanguageModelOptions
 >;
-
-// Vertex API requires another service tier format.
-export const VertexServiceTierMap = {
-  standard: 'SERVICE_TIER_STANDARD',
-  flex: 'SERVICE_TIER_FLEX',
-  priority: 'SERVICE_TIER_PRIORITY',
-} as const;

package/src/google-provider.ts

@@ -89,13 +89,17 @@ export interface GoogleGenerativeAIProvider extends ProviderV3 {
 
   /**
    * Creates a language model targeting the Gemini Interactions API
-   * (`POST /v1beta/interactions`). Pass either a model ID (string) or
-   * `{ agent: <name> }` to use a Gemini agent preset.
+   * (`POST /v1beta/interactions`). Pass:
+   *   - a model ID (string),
+   *   - `{ agent: <name> }` to use a known Gemini agent preset, or
+   *   - `{ managedAgent: <name> }` to use a user-defined agent created via
+   *     the `/v1beta/agents` endpoint.
    */
   interactions(
     modelIdOrAgent:
       | GoogleInteractionsModelId
-      | { agent: GoogleInteractionsAgentName },
+      | { agent: GoogleInteractionsAgentName }
+      | { managedAgent: string },
   ): LanguageModelV3;
 
   tools: typeof googleTools;
@@ -214,7 +218,8 @@ export function createGoogleGenerativeAI(
   const createInteractionsModel = (
     modelIdOrAgent:
       | GoogleInteractionsModelId
-      | { agent: GoogleInteractionsAgentName },
+      | { agent: GoogleInteractionsAgentName }
+      | { managedAgent: string },
   ) =>
     new GoogleInteractionsLanguageModel(
       modelIdOrAgent as GoogleInteractionsModelInput,