@ai-sdk/google 3.0.49 → 3.0.51

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

@@ -74,14 +74,24 @@ export class GoogleGenerativeAIEmbeddingModel implements EmbeddingModelV3 {
       headers,
     );
 
-    const multimodalContent = googleOptions?.content ?? [];
+    const multimodalContent = googleOptions?.content;
+
+    if (
+      multimodalContent != null &&
+      multimodalContent.length !== values.length
+    ) {
+      throw new Error(
+        `The number of multimodal content entries (${multimodalContent.length}) must match the number of values (${values.length}).`,
+      );
+    }
 
     // For single embeddings, use the single endpoint
     if (values.length === 1) {
+      const valueParts = multimodalContent?.[0];
       const textPart = values[0] ? [{ text: values[0] }] : [];
       const parts =
-        multimodalContent.length > 0
-          ? [...textPart, ...multimodalContent]
+        valueParts != null
+          ? [...textPart, ...valueParts]
           : [{ text: values[0] }];
 
       const {
@@ -116,7 +126,6 @@ export class GoogleGenerativeAIEmbeddingModel implements EmbeddingModelV3 {
     }
 
     // For multiple values, use the batch endpoint
-    // If multimodal content is provided, merge it into each request's parts
     const {
       responseHeaders,
       value: response,
@@ -125,15 +134,16 @@ export class GoogleGenerativeAIEmbeddingModel implements EmbeddingModelV3 {
       url: `${this.config.baseURL}/models/${this.modelId}:batchEmbedContents`,
       headers: mergedHeaders,
       body: {
-        requests: values.map(value => {
+        requests: values.map((value, index) => {
+          const valueParts = multimodalContent?.[index];
           const textPart = value ? [{ text: value }] : [];
           return {
             model: `models/${this.modelId}`,
             content: {
               role: 'user',
               parts:
-                multimodalContent.length > 0
-                  ? [...textPart, ...multimodalContent]
+                valueParts != null
+                  ? [...textPart, ...valueParts]
                   : [{ text: value }],
             },
             outputDimensionality: googleOptions?.outputDimensionality,

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

@@ -7,6 +7,7 @@ import { z } from 'zod/v4';
 
 export type GoogleGenerativeAIEmbeddingModelId =
   | 'gemini-embedding-001'
+  | 'gemini-embedding-2-preview'
   | (string & {});
 
 const googleEmbeddingContentPartSchema = z.union([
@@ -54,11 +55,17 @@ export const googleEmbeddingModelOptions = lazySchema(() =>
         .optional(),
 
       /**
-       * Optional. Multimodal content parts for embedding non-text content
-       * (images, video, PDF, audio). When provided, these parts are merged
-       * with the text values in the embedding request.
+       * Optional. Per-value multimodal content parts for embedding non-text
+       * content (images, video, PDF, audio). Each entry corresponds to the
+       * embedding value at the same index and its parts are merged with the
+       * text value in the request. Use `null` for entries that are text-only.
+       *
+       * The array length must match the number of values being embedded. In
+       * the case of a single embedding, the array length must be 1.
        */
-      content: z.array(googleEmbeddingContentPartSchema).min(1).optional(),
+      content: z
+        .array(z.array(googleEmbeddingContentPartSchema).min(1).nullable())
+        .optional(),
     }),
   ),
 );

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

@@ -38,7 +38,10 @@ import {
   GoogleGenerativeAIModelId,
   googleLanguageModelOptions,
 } from './google-generative-ai-options';
-import { GoogleGenerativeAIContentPart } from './google-generative-ai-prompt';
+import {
+  GoogleGenerativeAIContentPart,
+  GoogleGenerativeAIProviderMetadata,
+} from './google-generative-ai-prompt';
 import { prepareTools } from './google-prepare-tools';
 import { mapGoogleGenerativeAIFinishReason } from './map-google-generative-ai-finish-reason';
 
@@ -355,7 +358,8 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
           urlContextMetadata: candidate.urlContextMetadata ?? null,
           safetyRatings: candidate.safetyRatings ?? null,
           usageMetadata: usageMetadata ?? null,
-        },
+          finishMessage: candidate.finishMessage ?? null,
+        } satisfies GoogleGenerativeAIProviderMetadata,
       },
       request: { body: args },
       response: {
@@ -674,16 +678,10 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
                   groundingMetadata: lastGroundingMetadata,
                   urlContextMetadata: lastUrlContextMetadata,
                   safetyRatings: candidate.safetyRatings ?? null,
-                },
+                  usageMetadata: usageMetadata ?? null,
+                  finishMessage: candidate.finishMessage ?? null,
+                } satisfies GoogleGenerativeAIProviderMetadata,
               };
-              if (usageMetadata != null) {
-                (
-                  providerMetadata[providerOptionsName] as Record<
-                    string,
-                    unknown
-                  >
-                ).usageMetadata = usageMetadata;
-              }
             }
           },
 
@@ -1012,6 +1010,7 @@ const responseSchema = lazySchema(() =>
         z.object({
           content: getContentSchema().nullish().or(z.object({}).strict()),
           finishReason: z.string().nullish(),
+          finishMessage: z.string().nullish(),
           safetyRatings: z.array(getSafetyRatingSchema()).nullish(),
           groundingMetadata: getGroundingMetadataSchema().nullish(),
           urlContextMetadata: getUrlContextMetadataSchema().nullish(),
@@ -1047,6 +1046,14 @@ export type SafetyRatingSchema = NonNullable<
   InferSchema<typeof responseSchema>['candidates'][number]['safetyRatings']
 >[number];
 
+export type PromptFeedbackSchema = NonNullable<
+  InferSchema<typeof responseSchema>['promptFeedback']
+>;
+
+export type UsageMetadataSchema = NonNullable<
+  InferSchema<typeof responseSchema>['usageMetadata']
+>;
+
 // limited version of the schema, focussed on what is needed for the implementation
 // this approach limits breakages when the API changes and increases efficiency
 const chunkSchema = lazySchema(() =>
@@ -1057,6 +1064,7 @@ const chunkSchema = lazySchema(() =>
           z.object({
             content: getContentSchema().nullish(),
             finishReason: z.string().nullish(),
+            finishMessage: z.string().nullish(),
             safetyRatings: z.array(getSafetyRatingSchema()).nullish(),
             groundingMetadata: getGroundingMetadataSchema().nullish(),
             urlContextMetadata: getUrlContextMetadataSchema().nullish(),

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

@@ -1,6 +1,8 @@
 import {
   GroundingMetadataSchema,
+  PromptFeedbackSchema,
   UrlContextMetadataSchema,
+  UsageMetadataSchema,
 } from './google-generative-ai-language-model';
 import { type SafetyRatingSchema } from './google-generative-ai-language-model';
 
@@ -31,8 +33,15 @@ export type GoogleGenerativeAIUrlContextMetadata = UrlContextMetadataSchema;
 
 export type GoogleGenerativeAISafetyRating = SafetyRatingSchema;
 
+export type GoogleGenerativeAIPromptFeedback = PromptFeedbackSchema;
+
+export type GoogleGenerativeAIUsageMetadata = UsageMetadataSchema;
+
 export interface GoogleGenerativeAIProviderMetadata {
+  promptFeedback: GoogleGenerativeAIPromptFeedback | null;
   groundingMetadata: GoogleGenerativeAIGroundingMetadata | null;
   urlContextMetadata: GoogleGenerativeAIUrlContextMetadata | null;
   safetyRatings: GoogleGenerativeAISafetyRating[] | null;
+  usageMetadata: GoogleGenerativeAIUsageMetadata | null;
+  finishMessage: string | null;
 }