@ai-sdk/google 3.0.47 → 3.0.48

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/google",
-  "version": "3.0.47",
+  "version": "3.0.48",
   "license": "Apache-2.0",
   "sideEffects": false,
   "main": "./dist/index.js",
@@ -44,8 +44,8 @@
     "tsup": "^8",
     "typescript": "5.8.3",
     "zod": "3.25.76",
-    "@ai-sdk/test-server": "1.0.3",
-    "@vercel/ai-tsconfig": "0.0.0"
+    "@vercel/ai-tsconfig": "0.0.0",
+    "@ai-sdk/test-server": "1.0.3"
   },
   "peerDependencies": {
     "zod": "^3.25.76 || ^4.1.8"

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

@@ -74,8 +74,16 @@ export class GoogleGenerativeAIEmbeddingModel implements EmbeddingModelV3 {
       headers,
     );
 
-    // For single embeddings, use the single endpoint (ratelimits, etc.)
+    const multimodalContent = googleOptions?.content ?? [];
+
+    // For single embeddings, use the single endpoint
     if (values.length === 1) {
+      const textPart = values[0] ? [{ text: values[0] }] : [];
+      const parts =
+        multimodalContent.length > 0
+          ? [...textPart, ...multimodalContent]
+          : [{ text: values[0] }];
+
       const {
         responseHeaders,
         value: response,
@@ -86,7 +94,7 @@ export class GoogleGenerativeAIEmbeddingModel implements EmbeddingModelV3 {
         body: {
           model: `models/${this.modelId}`,
           content: {
-            parts: [{ text: values[0] }],
+            parts,
           },
           outputDimensionality: googleOptions?.outputDimensionality,
           taskType: googleOptions?.taskType,
@@ -107,6 +115,8 @@ 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,
@@ -115,12 +125,21 @@ export class GoogleGenerativeAIEmbeddingModel implements EmbeddingModelV3 {
       url: `${this.config.baseURL}/models/${this.modelId}:batchEmbedContents`,
       headers: mergedHeaders,
       body: {
-        requests: values.map(value => ({
-          model: `models/${this.modelId}`,
-          content: { role: 'user', parts: [{ text: value }] },
-          outputDimensionality: googleOptions?.outputDimensionality,
-          taskType: googleOptions?.taskType,
-        })),
+        requests: values.map(value => {
+          const textPart = value ? [{ text: value }] : [];
+          return {
+            model: `models/${this.modelId}`,
+            content: {
+              role: 'user',
+              parts:
+                multimodalContent.length > 0
+                  ? [...textPart, ...multimodalContent]
+                  : [{ text: value }],
+            },
+            outputDimensionality: googleOptions?.outputDimensionality,
+            taskType: googleOptions?.taskType,
+          };
+        }),
       },
       failedResponseHandler: googleFailedResponseHandler,
       successfulResponseHandler: createJsonResponseHandler(

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

@@ -9,6 +9,16 @@ export type GoogleGenerativeAIEmbeddingModelId =
   | 'gemini-embedding-001'
   | (string & {});
 
+const googleEmbeddingContentPartSchema = z.union([
+  z.object({ text: z.string() }),
+  z.object({
+    inlineData: z.object({
+      mimeType: z.string(),
+      data: z.string(),
+    }),
+  }),
+]);
+
 export const googleEmbeddingModelOptions = lazySchema(() =>
   zodSchema(
     z.object({
@@ -42,6 +52,13 @@ export const googleEmbeddingModelOptions = lazySchema(() =>
           'CODE_RETRIEVAL_QUERY',
         ])
         .optional(),
+
+      /**
+       * Optional. Multimodal content parts for embedding non-text content
+       * (images, video, PDF, audio). When provided, these parts are used
+       * instead of the text values in the embedding request.
+       */
+      content: z.array(googleEmbeddingContentPartSchema).min(1).optional(),
     }),
   ),
 );