@ai-sdk/google 4.0.107 → 4.0.109

package/docs/15-google.mdx

@@ -97,7 +97,11 @@ You can use the following optional settings to customize the Google provider ins
   Use a different URL prefix for API calls, e.g. to use proxy servers.
   The default prefix is `https://generativelanguage.googleapis.com/v1beta`.
 
-- **apiKey** _string_
+- **apiKey** _string | PromiseLike<string> | (() => string | PromiseLike<string>)_
+
+  Pass a function to resolve credentials before each request, including when keys
+  rotate. The function can return a string or a promise. Credential caching and
+  refresh are managed by the function.
 
   API key that is being sent using the `x-goog-api-key` header.
   It defaults to the `GOOGLE_GENERATIVE_AI_API_KEY` environment variable.
@@ -540,7 +544,14 @@ const result = await generateText({
 });
 ```
 
-You can also use YouTube URLs directly:
+You can also use YouTube URLs directly. Pass `videoMetadata` through
+`providerOptions.google` on a video file part to clip a segment or control the
+frame rate. This also works with inline video data and uploaded file references.
+The metadata supports these optional fields:
+
+- `startOffset` (string): Start time as a duration in seconds, e.g. `'60s'` or `'3.5s'`.
+- `endOffset` (string): End time in the same duration format.
+- `fps` (number): Frames per second in the range `(0, 24]`. The default is `1`.
 
 ```ts
 import { google } from '@ai-sdk/google';
@@ -554,12 +565,21 @@ const result = await generateText({
       content: [
         {
           type: 'text',
-          text: 'Summarize this video',
+          text: 'Summarize this part of the video',
         },
         {
           type: 'file',
           data: 'https://www.youtube.com/watch?v=dQw4w9WgXcQ',
           mediaType: 'video/mp4',
+          providerOptions: {
+            google: {
+              videoMetadata: {
+                startOffset: '60s',
+                endOffset: '600s',
+                fps: 1,
+              },
+            },
+          },
         },
       ],
     },
@@ -1249,6 +1269,8 @@ will still validate the response against the schema in `Output.object`.
 
 ### Model Capabilities
 
+{/* model-capabilities:google-language:start */}
+
 | Model                                 | Image Input | Object Generation | Tool Usage | Tool Streaming | Google Search | URL Context |
 | ------------------------------------- | ----------- | ----------------- | ---------- | -------------- | ------------- | ----------- |
 | `gemini-nano-banana-2.1`              | <Check />   | <Cross />         | <Cross />  | <Cross />      | <Check />     | <Cross />   |
@@ -1270,6 +1292,8 @@ will still validate the response against the schema in `Output.object`.
 | `gemini-2.5-flash-lite-preview-06-17` | <Check />   | <Check />         | <Check />  | <Check />      | <Check />     | <Check />   |
 | `gemini-2.0-flash`                    | <Check />   | <Check />         | <Check />  | <Check />      | <Check />     | <Check />   |
 
+{/* model-capabilities:google-language:end */}
+
 <Note>
   The table above lists popular models. Please see the [Google Generative AI
   docs](https://ai.google.dev/gemini-api/docs/models/) for a full list of

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/google",
-  "version": "4.0.107",
+  "version": "4.0.109",
   "type": "module",
   "license": "Apache-2.0",
   "sideEffects": false,
@@ -35,8 +35,8 @@
     }
   },
   "dependencies": {
-    "@ai-sdk/provider": "4.0.34",
-    "@ai-sdk/provider-utils": "5.0.70"
+    "@ai-sdk/provider": "4.0.35",
+    "@ai-sdk/provider-utils": "5.0.72"
   },
   "devDependencies": {
     "@ai-sdk/test-server": "2.0.4",

package/src/convert-to-google-messages.ts

@@ -14,11 +14,12 @@ import {
   resolveProviderReference,
   secureJsonParse,
 } from '@ai-sdk/provider-utils';
-import type {
-  GoogleContent,
-  GoogleContentPart,
-  GoogleFunctionResponsePart,
-  GooglePrompt,
+import {
+  googleVideoMetadataSchema,
+  type GoogleContent,
+  type GoogleContentPart,
+  type GoogleFunctionResponsePart,
+  type GooglePrompt,
 } from './google-prompt';
 import {
   codeExecutionInputSchema,
@@ -330,6 +331,12 @@ export function convertToGoogleMessages(
             }
 
             case 'file': {
+              const rawVideoMetadata = readProviderOpts(part)?.videoMetadata;
+              const videoMetadata =
+                rawVideoMetadata == null
+                  ? undefined
+                  : googleVideoMetadataSchema.parse(rawVideoMetadata);
+
               switch (part.data.type) {
                 case 'url': {
                   parts.push({
@@ -341,6 +348,7 @@ export function convertToGoogleMessages(
                           ? part.data.originalUrl
                           : part.data.url.toString(),
                     },
+                    ...(videoMetadata != null ? { videoMetadata } : {}),
                   });
                   break;
                 }
@@ -359,6 +367,7 @@ export function convertToGoogleMessages(
                         provider: 'google',
                       }),
                     },
+                    ...(videoMetadata != null ? { videoMetadata } : {}),
                   });
                   break;
                 }
@@ -372,6 +381,7 @@ export function convertToGoogleMessages(
                         new TextEncoder().encode(part.data.text),
                       ),
                     },
+                    ...(videoMetadata != null ? { videoMetadata } : {}),
                   });
                   break;
                 }
@@ -381,6 +391,7 @@ export function convertToGoogleMessages(
                       mimeType: resolveFullMediaType({ part }),
                       data: convertToBase64(part.data.data),
                     },
+                    ...(videoMetadata != null ? { videoMetadata } : {}),
                   });
                   break;
                 }

package/src/google-batch.ts

@@ -292,7 +292,6 @@ export class GoogleBatch implements BatchV4<GoogleBatchModelIds> {
       }
     }
 
-    const headers = await this.getHeaders(options.headers);
     const createUrl = `${this.batchConfig.baseURL}/${getModelPath(
       modelId,
     )}:batchGenerateContent`;
@@ -300,7 +299,7 @@ export class GoogleBatch implements BatchV4<GoogleBatchModelIds> {
     if (fileParts == null) {
       const { value: operation } = await postJsonToApi({
         url: createUrl,
-        headers,
+        headers: await this.getHeaders(options.headers),
         body: inlineBatchBody,
         failedResponseHandler: googleFailedResponseHandler,
         successfulResponseHandler: createJsonResponseHandler(
@@ -329,7 +328,7 @@ export class GoogleBatch implements BatchV4<GoogleBatchModelIds> {
 
     const { value: uploadUrl } = await postJsonToApi({
       url: `${this.getBaseOrigin()}/upload/v1beta/files`,
-      headers: combineHeaders(headers, {
+      headers: combineHeaders(await this.getHeaders(options.headers), {
         'X-Goog-Upload-Protocol': 'resumable',
         'X-Goog-Upload-Command': 'start',
         'X-Goog-Upload-Header-Content-Length': String(inputFile.size),
@@ -370,7 +369,7 @@ export class GoogleBatch implements BatchV4<GoogleBatchModelIds> {
 
     const { value: operation } = await postJsonToApi({
       url: createUrl,
-      headers,
+      headers: await this.getHeaders(options.headers),
       body: {
         batch: {
           displayName,

package/src/google-embedding-model.ts

@@ -3,6 +3,7 @@ import {
   type EmbeddingModelV4,
 } from '@ai-sdk/provider';
 import {
+  type Resolvable,
   combineHeaders,
   createJsonResponseHandler,
   type EmbeddingModelProviderOptionsTransformer,
@@ -26,7 +27,7 @@ import {
 type GoogleEmbeddingConfig = {
   provider: string;
   baseURL: string;
-  headers?: () => Record<string, string | undefined>;
+  headers?: Resolvable<Record<string, string | undefined>>;
   fetch?: FetchFunction;
 };
 

package/src/google-files.ts

@@ -6,6 +6,8 @@ import {
   type SharedV4Warning,
 } from '@ai-sdk/provider';
 import {
+  resolve,
+  type Resolvable,
   combineHeaders,
   convertInlineFileDataToUint8Array,
   createJsonResponseHandler,
@@ -31,7 +33,7 @@ export type GoogleFilesUploadOptions = {
 interface GoogleFilesConfig {
   provider: string;
   baseURL: string;
-  headers: () => Record<string, string | undefined>;
+  headers: Resolvable<Record<string, string | undefined>>;
   fetch?: FetchFunction;
 }
 
@@ -53,10 +55,8 @@ export class GoogleFiles implements FilesV4 {
       schema: googleFilesUploadOptionsSchema,
     })) as GoogleFilesUploadOptions | undefined;
 
-    const resolvedHeaders = combineHeaders(
-      this.config.headers(),
-      options.headers,
-    );
+    const getHeaders = async () =>
+      combineHeaders(await resolve(this.config.headers), options.headers);
     const fetchFn = this.config.fetch ?? globalThis.fetch;
 
     const warnings: Array<SharedV4Warning> = [];
@@ -74,7 +74,7 @@ export class GoogleFiles implements FilesV4 {
     const initResponse = await fetchFn(`${baseOrigin}/upload/v1beta/files`, {
       method: 'POST',
       headers: {
-        ...resolvedHeaders,
+        ...(await getHeaders()),
         'X-Goog-Upload-Protocol': 'resumable',
         'X-Goog-Upload-Command': 'start',
         'X-Goog-Upload-Header-Content-Length': String(fileBytes.length),
@@ -152,7 +152,7 @@ export class GoogleFiles implements FilesV4 {
       const { value: fileStatus } = await getFromApi({
         url: `${this.config.baseURL}/${filePath}`,
         validateUrl: false,
-        headers: combineHeaders(resolvedHeaders),
+        headers: await getHeaders(),
         successfulResponseHandler: createJsonResponseHandler(
           googleFileResponseSchema,
         ),

package/src/google-json-accumulator.ts

@@ -74,6 +74,9 @@ export class GoogleJSONAccumulator {
           (existingValue as string) + arg.stringValue,
         );
         delta += escaped;
+        if (arg.willContinue !== true) {
+          delta += this.closeOpenString();
+        }
         continue;
       }
 
@@ -81,8 +84,15 @@ export class GoogleJSONAccumulator {
       if (resolved == null) {
         // A value-less entry for a path without a value is an empty container
         // (an empty container has no leaves to stream). Value-less entries for
-        // existing paths are string or array terminators and are ignored.
-        if (existingValue !== undefined || arg.willContinue === true) continue;
+        // existing paths are string or array terminators. A string terminator
+        // closes the open string, an array terminator is ignored.
+        if (existingValue !== undefined) {
+          if (typeof existingValue === 'string' && arg.willContinue !== true) {
+            delta += this.closeOpenString();
+          }
+          continue;
+        }
+        if (arg.willContinue === true) continue;
 
         setNestedValue(
           this.accumulatedArgs,
@@ -115,6 +125,18 @@ export class GoogleJSONAccumulator {
     return { finalJSON: finalArgs, closingDelta };
   }
 
+  /**
+   * Input: stringOpen=true
+   * Output: '"' (and stringOpen=false), or '' when no string is open
+   */
+  private closeOpenString(): string {
+    if (!this.stringOpen) {
+      return '';
+    }
+    this.stringOpen = false;
+    return '"';
+  }
+
   /**
    * Input: pathStack=[] (first call) or pathStack=[root,...] (subsequent calls)
    * Output: '{' (first call) or '' (subsequent calls)
@@ -139,12 +161,7 @@ export class GoogleJSONAccumulator {
     arg: PartialArg,
     valueJson: string,
   ): string {
-    let fragment = '';
-
-    if (this.stringOpen) {
-      fragment += '"';
-      this.stringOpen = false;
-    }
+    let fragment = this.closeOpenString();
 
     fragment += this.ensureRoot();
 
@@ -172,12 +189,7 @@ export class GoogleJSONAccumulator {
     targetSegments: PathSegment[],
     isArray: boolean,
   ): string {
-    let fragment = '';
-
-    if (this.stringOpen) {
-      fragment += '"';
-      this.stringOpen = false;
-    }
+    let fragment = this.closeOpenString();
 
     fragment += this.ensureRoot();
 

package/src/google-language-model.ts

@@ -58,7 +58,7 @@ import {
   type GoogleLanguageModelOptions,
   type GoogleModelId,
 } from './google-language-model-options';
-import { getGoogleModelCapabilities } from './google-model-capabilities';
+import { getGoogleLanguageModelEntry } from './google-model-catalog';
 import type { GoogleProviderMetadata } from './google-prompt';
 import { prepareTools } from './google-prepare-tools';
 import {
@@ -74,8 +74,6 @@ const configurableSafetySettingCategories = [
   'HARM_CATEGORY_SEXUALLY_EXPLICIT',
 ] as const;
 
-const gemini25ModelPattern = /(^|[/.])gemini-2\.5(?:[.-]|$)/i;
-
 const googleCloudStorageFunctionResponseUrls = {
   'image/png': [/^gs:\/\/.*$/],
   'image/jpeg': [/^gs:\/\/.*$/],
@@ -296,9 +294,12 @@ export class GoogleLanguageModel implements LanguageModelV4 {
       }
     }
 
-    const isGemmaModel = modelId.toLowerCase().startsWith('gemma-');
+    const entry = getGoogleLanguageModelEntry(modelId);
+    const isGemmaModel =
+      entry.requestConfig.language.systemInstruction === 'user';
     const isGemini25DeveloperApiModel =
-      !isVertexProvider && gemini25ModelPattern.test(modelId);
+      !isVertexProvider &&
+      entry.capabilities.language.sampling.penalties !== true;
 
     if (isGemini25DeveloperApiModel && frequencyPenalty != null) {
       warnings.push({
@@ -313,7 +314,8 @@ export class GoogleLanguageModel implements LanguageModelV4 {
       });
     }
 
-    const { usesGemini3Features } = getGoogleModelCapabilities(modelId);
+    const usesGemini3Features =
+      entry.capabilities.language.tools.functionResponseParts;
     const supportedFunctionResponseUrls =
       usesGemini3Features &&
       config.downloadToolResultFiles?.supportsGoogleCloudStorageUrls
@@ -466,6 +468,8 @@ export class GoogleLanguageModel implements LanguageModelV4 {
     { isStreaming = false }: { isStreaming?: boolean } = {},
   ) {
     const { responseFormat, tools, toolChoice } = options;
+    const entry = getGoogleLanguageModelEntry(this.modelId);
+    const { structuredOutput } = entry.capabilities.language;
     let jsonResponseTool: LanguageModelV4FunctionTool | undefined;
     // Gemini models before Gemini 3 reject JSON response formatting together
     // with function tools. Gemini 3 supports the combination only with automatic
@@ -474,8 +478,8 @@ export class GoogleLanguageModel implements LanguageModelV4 {
       responseFormat?.type === 'json' &&
       responseFormat.schema != null &&
       tools?.some(tool => tool.type === 'function') &&
-      /(^|\/)gemini-/i.test(this.modelId) &&
-      (!getGoogleModelCapabilities(this.modelId).usesGemini3Features ||
+      entry.requestConfig.language.jsonResponseTool &&
+      (!structuredOutput.withFunctionTools ||
         toolChoice?.type === 'required' ||
         toolChoice?.type === 'tool')
     ) {
@@ -754,18 +758,16 @@ export class GoogleLanguageModel implements LanguageModelV4 {
       jsonResponseToolName,
     } = await this.getArgs(options);
 
-    const mergedHeaders = combineHeaders(
-      this.config.headers ? await resolve(this.config.headers) : undefined,
-      options.headers,
-      extraHeaders,
-    );
-
-    const generateContent = (body: typeof args) =>
+    const generateContent = async (body: typeof args) =>
       postJsonToApi({
         url: `${this.config.baseURL}/${getModelPath(
           this.modelId,
         )}:generateContent`,
-        headers: mergedHeaders,
+        headers: combineHeaders(
+          await resolve(this.config.headers),
+          options.headers,
+          extraHeaders,
+        ),
         body,
         failedResponseHandler: googleFailedResponseHandler,
         successfulResponseHandler: createJsonResponseHandler(responseSchema),
@@ -875,18 +877,16 @@ export class GoogleLanguageModel implements LanguageModelV4 {
         providerOptionsNames.map(name => [name, payload]),
       ) as SharedV4ProviderMetadata;
 
-    const headers = combineHeaders(
-      this.config.headers ? await resolve(this.config.headers) : undefined,
-      options.headers,
-      extraHeaders,
-    );
-
-    const streamGenerateContent = (body: typeof args) =>
+    const streamGenerateContent = async (body: typeof args) =>
       postJsonToApi({
         url: `${this.config.baseURL}/${getModelPath(
           this.modelId,
         )}:streamGenerateContent?alt=sse`,
-        headers,
+        headers: combineHeaders(
+          await resolve(this.config.headers),
+          options.headers,
+          extraHeaders,
+        ),
         body,
         failedResponseHandler: googleFailedResponseHandler,
         successfulResponseHandler:
@@ -1571,18 +1571,6 @@ export class GoogleLanguageModel implements LanguageModelV4 {
   }
 }
 
-function getMaxOutputTokensForGemini25Model(): number {
-  return 65536;
-}
-
-function getMaxThinkingTokensForGemini25Model(modelId: string): number {
-  const id = modelId.toLowerCase();
-  if (id.includes('2.5-pro') || id.includes('gemini-3-pro-image')) {
-    return 32768;
-  }
-  return 24576;
-}
-
 type GoogleThinkingConfig = NonNullable<
   InferSchema<typeof googleLanguageModelOptions>['thinkingConfig']
 >;
@@ -1601,8 +1589,8 @@ function resolveThinkingConfig({
   }
 
   if (
-    getGoogleModelCapabilities(modelId).usesGemini3Features &&
-    !modelId.includes('gemini-3-pro-image')
+    getGoogleLanguageModelEntry(modelId).requestConfig.language
+      .thinkingFormat === 'level'
   ) {
     return resolveGemini3ThinkingConfig({ reasoning, modelId, warnings });
   }
@@ -1622,22 +1610,21 @@ function resolveGemini3ThinkingConfig({
   modelId: string;
   warnings: SharedV4Warning[];
 }): Pick<GoogleThinkingConfig, 'thinkingLevel'> | undefined {
-  const minimumThinkingLevel = getMinimumThinkingLevelForGemini3Model(modelId);
-
-  if (reasoning === 'none') {
-    // It's not possible to fully disable thinking with Gemini 3.
-    return { thinkingLevel: minimumThinkingLevel };
-  }
+  const minimumThinkingLevel =
+    getGoogleLanguageModelEntry(modelId).capabilities.language.reasoning
+      .efforts?.[0];
 
+  // It's not possible to fully disable thinking with Gemini 3, so reasoning
+  // 'none' maps to the lowest supported thinking level.
   const thinkingLevel = mapReasoningToProviderEffort({
     reasoning,
     effortMap: {
-      minimal: minimumThinkingLevel,
+      ...(minimumThinkingLevel === 'minimal' && {
+        minimal: 'minimal' as const,
+      }),
       low: 'low',
       medium: 'medium',
       high: 'high',
-      xhigh: 'high',
-      max: 'high',
     },
     warnings,
   });
@@ -1649,41 +1636,6 @@ function resolveGemini3ThinkingConfig({
   return { thinkingLevel };
 }
 
-function getMinimumThinkingLevelForGemini3Model(
-  modelId: string,
-): 'minimal' | 'low' {
-  const modelName = /(?:^|[/.])(gemini-[^/]*)$/i
-    .exec(modelId)?.[1]
-    .toLowerCase();
-
-  if (modelName === 'gemini-flash-latest') {
-    return 'low';
-  }
-
-  const proVersionMatch = /^gemini-(\d+)(?:\.\d+)?-pro(?:$|-)/.exec(
-    modelName ?? '',
-  );
-
-  if (Number(proVersionMatch?.[1]) >= 3) {
-    return 'low';
-  }
-
-  const versionMatch = /^gemini-(\d+)\.(\d+)-flash(?:$|-(?!lite(?:-|$)))/.exec(
-    modelName ?? '',
-  );
-
-  if (versionMatch == null) {
-    return 'minimal';
-  }
-
-  const majorVersion = Number(versionMatch[1]);
-  const minorVersion = Number(versionMatch[2]);
-
-  return majorVersion > 3 || (majorVersion === 3 && minorVersion >= 7)
-    ? 'low'
-    : 'minimal';
-}
-
 function resolveGemini25ThinkingConfig({
   reasoning,
   modelId,
@@ -1695,21 +1647,21 @@ function resolveGemini25ThinkingConfig({
   >;
   modelId: string;
   warnings: SharedV4Warning[];
-}): Pick<GoogleThinkingConfig, 'thinkingBudget'> | undefined {
-  if (reasoning === 'none') {
+}): Pick<GoogleThinkingConfig, 'thinkingBudget'> {
+  const entry = getGoogleLanguageModelEntry(modelId);
+  const minThinkingBudget = entry.capabilities.language.reasoning.minBudget;
+
+  if (reasoning === 'none' && minThinkingBudget === 0) {
     return { thinkingBudget: 0 };
   }
 
   const thinkingBudget = mapReasoningToProviderBudget({
     reasoning,
-    maxOutputTokens: getMaxOutputTokensForGemini25Model(),
-    maxReasoningBudget: getMaxThinkingTokensForGemini25Model(modelId),
-    minReasoningBudget: 0,
+    maxOutputTokens: entry.requestConfig.language.reasoningMaxOutputTokens,
+    maxReasoningBudget: entry.capabilities.language.reasoning.maxBudget,
+    minReasoningBudget: minThinkingBudget,
     warnings,
   });
-  if (thinkingBudget == null) {
-    return undefined;
-  }
   return { thinkingBudget };
 }
 

package/src/google-model-capabilities.ts

@@ -1,44 +1,11 @@
-type GoogleModelCapabilities = {
-  supportsGemini2Tools: boolean;
-  supportsFileSearch: boolean;
-  usesGemini3Features: boolean;
-};
-
-const gemini1ModelPattern = /(^|[/.])gemini-1(?:[.-]|$)/i;
-const gemini2ModelPattern = /(^|[/.])gemini-2(?:[.-]|$)/i;
-const gemini25ModelPattern = /(^|[/.])gemini-2\.5(?:[.-]|$)/i;
-const geminiModelPattern = /(^|[/.])gemini-/i;
-
-function isKnownPreGemini2Model(modelId: string): boolean {
-  return (
-    gemini1ModelPattern.test(modelId) ||
-    /(^|[/.])gemini-pro(?:-vision)?$/i.test(modelId) ||
-    /(^|[/.])gemini-robotics-er-1\.5(?:[.-]|$)/i.test(modelId)
-  );
-}
-
-/**
- * Classifies Gemini capabilities by excluding known older generations.
- *
- * Google model IDs are open-ended, so unrecognized Gemini IDs and aliases
- * intentionally inherit the newest supported behavior. Add exceptions here
- * only when a model is known to require a legacy request shape.
- */
-export function getGoogleModelCapabilities(
-  modelId: string,
-): GoogleModelCapabilities {
-  const isGeminiModel = geminiModelPattern.test(modelId);
-  const isGemini2Model = gemini2ModelPattern.test(modelId);
-  const isKnownPreGemini2 = isKnownPreGemini2Model(modelId);
-  const isKnownOlderModel = isKnownPreGemini2 || isGemini2Model;
-  const usesGemini3Features = isGeminiModel && !isKnownOlderModel;
+import { getGoogleLanguageModelEntry } from './google-model-catalog';
 
+/** Compatibility projection of the resolved catalog for existing request code. */
+export function getGoogleModelCapabilities(modelId: string) {
+  const { tools } = getGoogleLanguageModelEntry(modelId).capabilities.language;
   return {
-    supportsGemini2Tools:
-      (isGeminiModel && !isKnownPreGemini2) ||
-      modelId.toLowerCase().includes('nano-banana'),
-    supportsFileSearch:
-      gemini25ModelPattern.test(modelId) || usesGemini3Features,
-    usesGemini3Features,
+    supportsGemini2Tools: tools.gemini2,
+    supportsFileSearch: tools.fileSearch,
+    usesGemini3Features: tools.mixed,
   };
 }