@ai-sdk/google 3.0.130 → 3.0.132

package/CHANGELOG.md

@@ -1,5 +1,20 @@
 # @ai-sdk/google
 
+## 3.0.132
+
+### Patch Changes
+
+- 0a2b345: feat(google): add `gemini-4-argon` model id and long decode support for `generateContent`: requests that reach the per-request cap (`CONTINUATION`) are continued automatically with usage summed across requests and the first request's prompt token count exposed as `providerMetadata.google.initialPromptTokenCount`, and the `continuationToken` provider option / provider metadata resume responses that used up their `maxOutputTokens` budget. `generateText` with the Interactions API now polls `background: true` model calls until they finish.
+- Updated dependencies [604b4d3]
+  - @ai-sdk/provider-utils@4.0.58
+
+## 3.0.131
+
+### Patch Changes
+
+- fed6d78: feat(google): preserve file search custom metadata in retrieved context grounding chunks
+- 6dd6693: feat(google): add the 'gemini-nano-banana-2.1' model ID
+
 ## 3.0.130
 
 ### Patch Changes

package/dist/index.d.mts

@@ -13,7 +13,7 @@ declare const googleErrorDataSchema: _ai_sdk_provider_utils.LazySchema<{
 }>;
 type GoogleErrorData = InferSchema<typeof googleErrorDataSchema>;
 
-type GoogleGenerativeAIModelId = 'gemini-2.0-flash' | 'gemini-2.0-flash-001' | 'gemini-2.0-flash-lite' | 'gemini-2.0-flash-lite-001' | 'gemini-2.5-pro' | 'gemini-2.5-flash' | 'gemini-2.5-flash-image' | 'gemini-2.5-flash-lite' | 'gemini-2.5-flash-preview-tts' | 'gemini-2.5-pro-preview-tts' | 'gemini-2.5-flash-native-audio-latest' | 'gemini-2.5-flash-native-audio-preview-09-2025' | 'gemini-2.5-flash-native-audio-preview-12-2025' | 'gemini-2.5-computer-use-preview-10-2025' | 'gemini-3-pro-preview' | 'gemini-3-pro-image-preview' | 'gemini-3-flash-preview' | 'gemini-3.1-pro-preview' | 'gemini-3.1-pro-preview-customtools' | 'gemini-3.1-flash-image-preview' | 'gemini-3.1-flash-lite-preview' | 'gemini-3.1-flash-tts-preview' | 'gemini-3.5-flash' | 'gemini-3.5-flash-lite' | 'gemini-3.6-flash' | 'gemini-3.7-flash' | 'gemini-3.8-flash' | 'gemini-pro-latest' | 'gemini-flash-latest' | 'gemini-flash-lite-latest' | 'deep-research-pro-preview-12-2025' | 'nano-banana-pro-preview' | 'aqa' | 'gemini-robotics-er-1.5-preview' | 'gemma-3-1b-it' | 'gemma-3-4b-it' | 'gemma-3n-e4b-it' | 'gemma-3n-e2b-it' | 'gemma-3-12b-it' | 'gemma-3-27b-it' | (string & {});
+type GoogleGenerativeAIModelId = 'gemini-2.0-flash' | 'gemini-2.0-flash-001' | 'gemini-2.0-flash-lite' | 'gemini-2.0-flash-lite-001' | 'gemini-2.5-pro' | 'gemini-2.5-flash' | 'gemini-2.5-flash-image' | 'gemini-2.5-flash-lite' | 'gemini-2.5-flash-preview-tts' | 'gemini-2.5-pro-preview-tts' | 'gemini-2.5-flash-native-audio-latest' | 'gemini-2.5-flash-native-audio-preview-09-2025' | 'gemini-2.5-flash-native-audio-preview-12-2025' | 'gemini-2.5-computer-use-preview-10-2025' | 'gemini-3-pro-preview' | 'gemini-3-pro-image-preview' | 'gemini-3-flash-preview' | 'gemini-3.1-pro-preview' | 'gemini-3.1-pro-preview-customtools' | 'gemini-3.1-flash-image-preview' | 'gemini-3.1-flash-lite-preview' | 'gemini-3.1-flash-tts-preview' | 'gemini-3.5-flash' | 'gemini-3.5-flash-lite' | 'gemini-3.6-flash' | 'gemini-3.7-flash' | 'gemini-3.8-flash' | 'gemini-4-argon' | 'gemini-nano-banana-2.1' | 'gemini-pro-latest' | 'gemini-flash-latest' | 'gemini-flash-lite-latest' | 'deep-research-pro-preview-12-2025' | 'nano-banana-pro-preview' | 'aqa' | 'gemini-robotics-er-1.5-preview' | 'gemma-3-1b-it' | 'gemma-3-4b-it' | 'gemma-3n-e4b-it' | 'gemma-3n-e2b-it' | 'gemma-3-12b-it' | 'gemma-3-27b-it' | (string & {});
 declare const googleLanguageModelOptions: _ai_sdk_provider_utils.LazySchema<{
     responseModalities?: ("TEXT" | "IMAGE")[] | undefined;
     thinkingConfig?: {
@@ -49,6 +49,7 @@ declare const googleLanguageModelOptions: _ai_sdk_provider_utils.LazySchema<{
     } | undefined;
     streamFunctionCallArguments?: boolean | undefined;
     serviceTier?: "standard" | "flex" | "priority" | undefined;
+    continuationToken?: string | undefined;
     sharedRequestType?: "standard" | "flex" | "priority" | undefined;
     requestType?: "shared" | undefined;
 }>;
@@ -111,6 +112,7 @@ declare const responseSchema: _ai_sdk_provider_utils.LazySchema<{
         } | null | undefined;
         finishReason?: string | null | undefined;
         finishMessage?: string | null | undefined;
+        continuationToken?: string | null | undefined;
         safetyRatings?: {
             category?: string | null | undefined;
             probability?: string | null | undefined;
@@ -142,6 +144,14 @@ declare const responseSchema: _ai_sdk_provider_utils.LazySchema<{
                     title?: string | null | undefined;
                     text?: string | null | undefined;
                     fileSearchStore?: string | null | undefined;
+                    customMetadata?: {
+                        key: string;
+                        stringValue?: string | null | undefined;
+                        numericValue?: number | null | undefined;
+                        stringListValue?: {
+                            values?: string[] | null | undefined;
+                        } | null | undefined;
+                    }[] | null | undefined;
                 } | null | undefined;
                 maps?: {
                     uri?: string | null | undefined;
@@ -236,9 +246,25 @@ interface GoogleGenerativeAIProviderMetadata {
     usageMetadata: GoogleGenerativeAIUsageMetadata | null;
     finishMessage: string | null;
     serviceTier: string | null;
+    /**
+     * Opaque token returned when a response stops with `MAX_TOKENS` because it
+     * used up its `maxOutputTokens` budget. Pass it back as
+     * `providerOptions.google.continuationToken` together with a higher
+     * `maxOutputTokens` and the original prompt to resume generation.
+     */
+    continuationToken?: string | null;
+    /**
+     * Prompt token count of the first request of a long decode response that
+     * was continued automatically (Gemini 4). Usage is summed across requests
+     * to match billing, and each continued request re-sends the prompt plus all
+     * output so far as input. Use this value instead of the summed input tokens
+     * for context window checks and compaction. Only present when the response
+     * spanned more than one request.
+     */
+    initialPromptTokenCount?: number | null;
 }
 
-type GoogleGenerativeAIImageModelId = 'imagen-4.0-generate-001' | 'imagen-4.0-ultra-generate-001' | 'imagen-4.0-fast-generate-001' | 'gemini-2.5-flash-image' | 'gemini-3-pro-image-preview' | 'gemini-3.1-flash-image-preview' | (string & {});
+type GoogleGenerativeAIImageModelId = 'imagen-4.0-generate-001' | 'imagen-4.0-ultra-generate-001' | 'imagen-4.0-fast-generate-001' | 'gemini-2.5-flash-image' | 'gemini-3-pro-image-preview' | 'gemini-3.1-flash-image-preview' | 'gemini-nano-banana-2.1' | (string & {});
 interface GoogleGenerativeAIImageSettings {
     /**
      * Override the maximum number of images per call (default 4)
@@ -306,7 +332,7 @@ type GoogleVideoModelOptions = {
  * the two surfaces (`:generateContent` vs `/interactions`) are independent and
  * may diverge over time.
  */
-type GoogleInteractionsModelId = 'gemini-2.5-computer-use-preview-10-2025' | 'gemini-2.5-flash' | 'gemini-2.5-flash-image' | 'gemini-2.5-flash-lite' | 'gemini-2.5-flash-lite-preview-09-2025' | 'gemini-2.5-flash-native-audio-preview-12-2025' | 'gemini-2.5-flash-preview-09-2025' | 'gemini-2.5-flash-preview-tts' | 'gemini-2.5-pro' | 'gemini-2.5-pro-preview-tts' | 'gemini-3-flash-preview' | 'gemini-3-pro-image-preview' | 'gemini-3-pro-preview' | 'gemini-3.1-pro-preview' | 'gemini-3.1-flash-image-preview' | 'gemini-3.1-flash-lite-preview' | 'gemini-3.1-flash-tts-preview' | 'gemini-3.5-flash' | 'gemini-3.5-flash-lite' | 'gemini-3.6-flash' | 'gemini-3.7-flash' | 'gemini-3.8-flash' | 'lyria-3-clip-preview' | 'lyria-3-pro-preview' | (string & {});
+type GoogleInteractionsModelId = 'gemini-2.5-computer-use-preview-10-2025' | 'gemini-2.5-flash' | 'gemini-2.5-flash-image' | 'gemini-2.5-flash-lite' | 'gemini-2.5-flash-lite-preview-09-2025' | 'gemini-2.5-flash-native-audio-preview-12-2025' | 'gemini-2.5-flash-preview-09-2025' | 'gemini-2.5-flash-preview-tts' | 'gemini-2.5-pro' | 'gemini-2.5-pro-preview-tts' | 'gemini-3-flash-preview' | 'gemini-3-pro-image-preview' | 'gemini-3-pro-preview' | 'gemini-3.1-pro-preview' | 'gemini-3.1-flash-image-preview' | 'gemini-3.1-flash-lite-preview' | 'gemini-3.1-flash-tts-preview' | 'gemini-3.5-flash' | 'gemini-3.5-flash-lite' | 'gemini-3.6-flash' | 'gemini-3.7-flash' | 'gemini-3.8-flash' | 'gemini-4-argon' | 'gemini-nano-banana-2.1' | 'lyria-3-clip-preview' | 'lyria-3-pro-preview' | (string & {});
 /**
  * Provider-options schema for `google.interactions(...)` calls. Read from the
  * shared `providerOptions.google.*` namespace (per PRD); per-call options that

package/dist/index.d.ts

@@ -13,7 +13,7 @@ declare const googleErrorDataSchema: _ai_sdk_provider_utils.LazySchema<{
 }>;
 type GoogleErrorData = InferSchema<typeof googleErrorDataSchema>;
 
-type GoogleGenerativeAIModelId = 'gemini-2.0-flash' | 'gemini-2.0-flash-001' | 'gemini-2.0-flash-lite' | 'gemini-2.0-flash-lite-001' | 'gemini-2.5-pro' | 'gemini-2.5-flash' | 'gemini-2.5-flash-image' | 'gemini-2.5-flash-lite' | 'gemini-2.5-flash-preview-tts' | 'gemini-2.5-pro-preview-tts' | 'gemini-2.5-flash-native-audio-latest' | 'gemini-2.5-flash-native-audio-preview-09-2025' | 'gemini-2.5-flash-native-audio-preview-12-2025' | 'gemini-2.5-computer-use-preview-10-2025' | 'gemini-3-pro-preview' | 'gemini-3-pro-image-preview' | 'gemini-3-flash-preview' | 'gemini-3.1-pro-preview' | 'gemini-3.1-pro-preview-customtools' | 'gemini-3.1-flash-image-preview' | 'gemini-3.1-flash-lite-preview' | 'gemini-3.1-flash-tts-preview' | 'gemini-3.5-flash' | 'gemini-3.5-flash-lite' | 'gemini-3.6-flash' | 'gemini-3.7-flash' | 'gemini-3.8-flash' | 'gemini-pro-latest' | 'gemini-flash-latest' | 'gemini-flash-lite-latest' | 'deep-research-pro-preview-12-2025' | 'nano-banana-pro-preview' | 'aqa' | 'gemini-robotics-er-1.5-preview' | 'gemma-3-1b-it' | 'gemma-3-4b-it' | 'gemma-3n-e4b-it' | 'gemma-3n-e2b-it' | 'gemma-3-12b-it' | 'gemma-3-27b-it' | (string & {});
+type GoogleGenerativeAIModelId = 'gemini-2.0-flash' | 'gemini-2.0-flash-001' | 'gemini-2.0-flash-lite' | 'gemini-2.0-flash-lite-001' | 'gemini-2.5-pro' | 'gemini-2.5-flash' | 'gemini-2.5-flash-image' | 'gemini-2.5-flash-lite' | 'gemini-2.5-flash-preview-tts' | 'gemini-2.5-pro-preview-tts' | 'gemini-2.5-flash-native-audio-latest' | 'gemini-2.5-flash-native-audio-preview-09-2025' | 'gemini-2.5-flash-native-audio-preview-12-2025' | 'gemini-2.5-computer-use-preview-10-2025' | 'gemini-3-pro-preview' | 'gemini-3-pro-image-preview' | 'gemini-3-flash-preview' | 'gemini-3.1-pro-preview' | 'gemini-3.1-pro-preview-customtools' | 'gemini-3.1-flash-image-preview' | 'gemini-3.1-flash-lite-preview' | 'gemini-3.1-flash-tts-preview' | 'gemini-3.5-flash' | 'gemini-3.5-flash-lite' | 'gemini-3.6-flash' | 'gemini-3.7-flash' | 'gemini-3.8-flash' | 'gemini-4-argon' | 'gemini-nano-banana-2.1' | 'gemini-pro-latest' | 'gemini-flash-latest' | 'gemini-flash-lite-latest' | 'deep-research-pro-preview-12-2025' | 'nano-banana-pro-preview' | 'aqa' | 'gemini-robotics-er-1.5-preview' | 'gemma-3-1b-it' | 'gemma-3-4b-it' | 'gemma-3n-e4b-it' | 'gemma-3n-e2b-it' | 'gemma-3-12b-it' | 'gemma-3-27b-it' | (string & {});
 declare const googleLanguageModelOptions: _ai_sdk_provider_utils.LazySchema<{
     responseModalities?: ("TEXT" | "IMAGE")[] | undefined;
     thinkingConfig?: {
@@ -49,6 +49,7 @@ declare const googleLanguageModelOptions: _ai_sdk_provider_utils.LazySchema<{
     } | undefined;
     streamFunctionCallArguments?: boolean | undefined;
     serviceTier?: "standard" | "flex" | "priority" | undefined;
+    continuationToken?: string | undefined;
     sharedRequestType?: "standard" | "flex" | "priority" | undefined;
     requestType?: "shared" | undefined;
 }>;
@@ -111,6 +112,7 @@ declare const responseSchema: _ai_sdk_provider_utils.LazySchema<{
         } | null | undefined;
         finishReason?: string | null | undefined;
         finishMessage?: string | null | undefined;
+        continuationToken?: string | null | undefined;
         safetyRatings?: {
             category?: string | null | undefined;
             probability?: string | null | undefined;
@@ -142,6 +144,14 @@ declare const responseSchema: _ai_sdk_provider_utils.LazySchema<{
                     title?: string | null | undefined;
                     text?: string | null | undefined;
                     fileSearchStore?: string | null | undefined;
+                    customMetadata?: {
+                        key: string;
+                        stringValue?: string | null | undefined;
+                        numericValue?: number | null | undefined;
+                        stringListValue?: {
+                            values?: string[] | null | undefined;
+                        } | null | undefined;
+                    }[] | null | undefined;
                 } | null | undefined;
                 maps?: {
                     uri?: string | null | undefined;
@@ -236,9 +246,25 @@ interface GoogleGenerativeAIProviderMetadata {
     usageMetadata: GoogleGenerativeAIUsageMetadata | null;
     finishMessage: string | null;
     serviceTier: string | null;
+    /**
+     * Opaque token returned when a response stops with `MAX_TOKENS` because it
+     * used up its `maxOutputTokens` budget. Pass it back as
+     * `providerOptions.google.continuationToken` together with a higher
+     * `maxOutputTokens` and the original prompt to resume generation.
+     */
+    continuationToken?: string | null;
+    /**
+     * Prompt token count of the first request of a long decode response that
+     * was continued automatically (Gemini 4). Usage is summed across requests
+     * to match billing, and each continued request re-sends the prompt plus all
+     * output so far as input. Use this value instead of the summed input tokens
+     * for context window checks and compaction. Only present when the response
+     * spanned more than one request.
+     */
+    initialPromptTokenCount?: number | null;
 }
 
-type GoogleGenerativeAIImageModelId = 'imagen-4.0-generate-001' | 'imagen-4.0-ultra-generate-001' | 'imagen-4.0-fast-generate-001' | 'gemini-2.5-flash-image' | 'gemini-3-pro-image-preview' | 'gemini-3.1-flash-image-preview' | (string & {});
+type GoogleGenerativeAIImageModelId = 'imagen-4.0-generate-001' | 'imagen-4.0-ultra-generate-001' | 'imagen-4.0-fast-generate-001' | 'gemini-2.5-flash-image' | 'gemini-3-pro-image-preview' | 'gemini-3.1-flash-image-preview' | 'gemini-nano-banana-2.1' | (string & {});
 interface GoogleGenerativeAIImageSettings {
     /**
      * Override the maximum number of images per call (default 4)
@@ -306,7 +332,7 @@ type GoogleVideoModelOptions = {
  * the two surfaces (`:generateContent` vs `/interactions`) are independent and
  * may diverge over time.
  */
-type GoogleInteractionsModelId = 'gemini-2.5-computer-use-preview-10-2025' | 'gemini-2.5-flash' | 'gemini-2.5-flash-image' | 'gemini-2.5-flash-lite' | 'gemini-2.5-flash-lite-preview-09-2025' | 'gemini-2.5-flash-native-audio-preview-12-2025' | 'gemini-2.5-flash-preview-09-2025' | 'gemini-2.5-flash-preview-tts' | 'gemini-2.5-pro' | 'gemini-2.5-pro-preview-tts' | 'gemini-3-flash-preview' | 'gemini-3-pro-image-preview' | 'gemini-3-pro-preview' | 'gemini-3.1-pro-preview' | 'gemini-3.1-flash-image-preview' | 'gemini-3.1-flash-lite-preview' | 'gemini-3.1-flash-tts-preview' | 'gemini-3.5-flash' | 'gemini-3.5-flash-lite' | 'gemini-3.6-flash' | 'gemini-3.7-flash' | 'gemini-3.8-flash' | 'lyria-3-clip-preview' | 'lyria-3-pro-preview' | (string & {});
+type GoogleInteractionsModelId = 'gemini-2.5-computer-use-preview-10-2025' | 'gemini-2.5-flash' | 'gemini-2.5-flash-image' | 'gemini-2.5-flash-lite' | 'gemini-2.5-flash-lite-preview-09-2025' | 'gemini-2.5-flash-native-audio-preview-12-2025' | 'gemini-2.5-flash-preview-09-2025' | 'gemini-2.5-flash-preview-tts' | 'gemini-2.5-pro' | 'gemini-2.5-pro-preview-tts' | 'gemini-3-flash-preview' | 'gemini-3-pro-image-preview' | 'gemini-3-pro-preview' | 'gemini-3.1-pro-preview' | 'gemini-3.1-flash-image-preview' | 'gemini-3.1-flash-lite-preview' | 'gemini-3.1-flash-tts-preview' | 'gemini-3.5-flash' | 'gemini-3.5-flash-lite' | 'gemini-3.6-flash' | 'gemini-3.7-flash' | 'gemini-3.8-flash' | 'gemini-4-argon' | 'gemini-nano-banana-2.1' | 'lyria-3-clip-preview' | 'lyria-3-pro-preview' | (string & {});
 /**
  * Provider-options schema for `google.interactions(...)` calls. Read from the
  * shared `providerOptions.google.*` namespace (per PRD); per-call options that

package/dist/index.js

@@ -235,7 +235,7 @@ var googleInteractionsTranscriptionResponseSchema = import_v43.z.object({
 var import_provider_utils25 = require("@ai-sdk/provider-utils");
 
 // src/version.ts
-var VERSION = true ? "3.0.130" : "0.0.0-test";
+var VERSION = true ? "3.0.132" : "0.0.0-test";
 
 // src/google-generative-ai-embedding-model.ts
 var import_provider2 = require("@ai-sdk/provider");
@@ -460,6 +460,53 @@ function validateMultimodalContentLength({
 var import_provider_utils8 = require("@ai-sdk/provider-utils");
 var import_v47 = require("zod/v4");
 
+// src/continue-google-stream.ts
+function continueGoogleStream({
+  stream,
+  getContinuationToken,
+  continueStream
+}) {
+  let reader = stream.getReader();
+  let pendingContinuationToken;
+  return new ReadableStream({
+    async pull(controller) {
+      while (true) {
+        const { done, value } = await reader.read();
+        if (!done) {
+          if (value.success) {
+            const token = getContinuationToken(value.value);
+            if (token != null) {
+              pendingContinuationToken = token;
+            }
+          }
+          controller.enqueue(value);
+          return;
+        }
+        if (pendingContinuationToken == null) {
+          controller.close();
+          return;
+        }
+        const continuationToken = pendingContinuationToken;
+        pendingContinuationToken = void 0;
+        try {
+          reader = (await continueStream(continuationToken)).getReader();
+        } catch (error) {
+          controller.enqueue({
+            success: false,
+            error,
+            rawValue: void 0
+          });
+          controller.close();
+          return;
+        }
+      }
+    },
+    async cancel(reason) {
+      await reader.cancel(reason);
+    }
+  });
+}
+
 // src/convert-google-generative-ai-usage.ts
 function convertGoogleGenerativeAIUsage(usage) {
   var _a2, _b, _c, _d;
@@ -498,6 +545,63 @@ function convertGoogleGenerativeAIUsage(usage) {
     raw: usage
   };
 }
+function addGoogleGenerativeAIUsageMetadata(a, b) {
+  if (a == null || b == null) {
+    return b != null ? b : a;
+  }
+  return {
+    ...a,
+    ...b,
+    promptTokenCount: addCounts(a.promptTokenCount, b.promptTokenCount),
+    candidatesTokenCount: addCounts(
+      a.candidatesTokenCount,
+      b.candidatesTokenCount
+    ),
+    toolUsePromptTokenCount: addCounts(
+      a.toolUsePromptTokenCount,
+      b.toolUsePromptTokenCount
+    ),
+    totalTokenCount: addCounts(a.totalTokenCount, b.totalTokenCount),
+    cachedContentTokenCount: addCounts(
+      a.cachedContentTokenCount,
+      b.cachedContentTokenCount
+    ),
+    thoughtsTokenCount: addCounts(a.thoughtsTokenCount, b.thoughtsTokenCount),
+    promptTokensDetails: addTokenDetails(
+      a.promptTokensDetails,
+      b.promptTokensDetails
+    ),
+    cacheTokensDetails: addTokenDetails(
+      a.cacheTokensDetails,
+      b.cacheTokensDetails
+    ),
+    candidatesTokensDetails: addTokenDetails(
+      a.candidatesTokensDetails,
+      b.candidatesTokensDetails
+    ),
+    toolUsePromptTokensDetails: addTokenDetails(
+      a.toolUsePromptTokensDetails,
+      b.toolUsePromptTokensDetails
+    )
+  };
+}
+function addCounts(a, b) {
+  return a == null && b == null ? b != null ? b : a : (a != null ? a : 0) + (b != null ? b : 0);
+}
+function addTokenDetails(a, b) {
+  var _a2;
+  if (a == null || b == null) {
+    return b != null ? b : a;
+  }
+  const tokenCounts = /* @__PURE__ */ new Map();
+  for (const { modality, tokenCount } of [...a, ...b]) {
+    tokenCounts.set(modality, ((_a2 = tokenCounts.get(modality)) != null ? _a2 : 0) + tokenCount);
+  }
+  return Array.from(tokenCounts, ([modality, tokenCount]) => ({
+    modality,
+    tokenCount
+  }));
+}
 
 // src/convert-json-schema-to-openapi-schema.ts
 var import_provider3 = require("@ai-sdk/provider");
@@ -1388,6 +1492,16 @@ var googleLanguageModelOptions = (0, import_provider_utils7.lazySchema)(
        * `serviceTier` body field. Gemini API only.
        */
       serviceTier: import_v46.z.enum(["standard", "flex", "priority"]).optional(),
+      /**
+       * Optional. Resumes a response that stopped with `MAX_TOKENS` because it
+       * used up its `maxOutputTokens` budget (Gemini 4 long decode). Pass the
+       * `continuationToken` from the previous response's provider metadata
+       * together with the original prompt and a higher `maxOutputTokens` (the
+       * budget is cumulative across requests). Requests that reach the
+       * per-request cap while budget remains (`CONTINUATION`) are continued
+       * automatically. Sent as the `continuationToken` body field.
+       */
+      continuationToken: import_v46.z.string().optional(),
       /**
        * Optional. Vertex AI only. Sent as the
        * `X-Vertex-AI-LLM-Shared-Request-Type` request header to select a
@@ -1936,6 +2050,9 @@ function mapGoogleGenerativeAIFinishReason({
     case "STOP":
       return hasToolCalls ? "tool-calls" : "stop";
     case "MAX_TOKENS":
+    // Gemini 4 long decode: the per-request cap was reached while budget
+    // remains. Continued automatically; only surfaced if continuing fails.
+    case "CONTINUATION":
       return "length";
     case "IMAGE_SAFETY":
     case "RECITATION":
@@ -2167,7 +2284,10 @@ var GoogleGenerativeAILanguageModel = class {
         toolConfig,
         cachedContent: googleOptions == null ? void 0 : googleOptions.cachedContent,
         labels: googleOptions == null ? void 0 : googleOptions.labels,
-        serviceTier: bodyServiceTier
+        serviceTier: bodyServiceTier,
+        ...(googleOptions == null ? void 0 : googleOptions.continuationToken) != null && {
+          continuationToken: googleOptions.continuationToken
+        }
       },
       warnings: [...warnings, ...toolWarnings],
       providerOptionsName,
@@ -2175,42 +2295,66 @@ var GoogleGenerativeAILanguageModel = class {
     };
   }
   async doGenerate(options) {
-    var _a2, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u;
+    var _a2, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v, _w, _x, _y, _z, _A;
     const { args, warnings, providerOptionsName, extraHeaders } = await this.getArgs(options);
     const mergedHeaders = (0, import_provider_utils8.combineHeaders)(
       await (0, import_provider_utils8.resolve)(this.config.headers),
       options.headers,
       extraHeaders
     );
-    const {
-      responseHeaders,
-      value: response,
-      rawValue: rawResponse
-    } = await (0, import_provider_utils8.postJsonToApi)({
+    const generateContent = (body) => (0, import_provider_utils8.postJsonToApi)({
       url: `${this.config.baseURL}/${getModelPath(
         this.modelId
       )}:generateContent`,
       headers: mergedHeaders,
-      body: args,
+      body,
       failedResponseHandler: googleFailedResponseHandler,
       successfulResponseHandler: (0, import_provider_utils8.createJsonResponseHandler)(responseSchema),
       abortSignal: options.abortSignal,
       fetch: this.config.fetch
     });
-    const candidate = (_a2 = response.candidates) == null ? void 0 : _a2[0];
-    const promptBlockReason = (_b = response.promptFeedback) == null ? void 0 : _b.blockReason;
+    let {
+      responseHeaders,
+      value: response,
+      rawValue: rawResponse
+    } = await generateContent(args);
+    const continuedParts = [];
+    let continuedUsageMetadata;
+    const firstPromptTokenCount = (_a2 = response.usageMetadata) == null ? void 0 : _a2.promptTokenCount;
+    while (true) {
+      const candidate2 = (_b = response.candidates) == null ? void 0 : _b[0];
+      const continuationToken = (candidate2 == null ? void 0 : candidate2.finishReason) === "CONTINUATION" ? candidate2.continuationToken : void 0;
+      if (continuationToken == null) {
+        break;
+      }
+      continuedParts.push(...(_d = (_c = candidate2 == null ? void 0 : candidate2.content) == null ? void 0 : _c.parts) != null ? _d : []);
+      continuedUsageMetadata = addGoogleGenerativeAIUsageMetadata(
+        continuedUsageMetadata,
+        (_e = response.usageMetadata) != null ? _e : void 0
+      );
+      ({
+        responseHeaders,
+        value: response,
+        rawValue: rawResponse
+      } = await generateContent({ ...args, continuationToken }));
+    }
+    const candidate = (_f = response.candidates) == null ? void 0 : _f[0];
+    const promptBlockReason = (_g = response.promptFeedback) == null ? void 0 : _g.blockReason;
     const confirmedPromptBlockReason = isConfirmedPromptBlockReason(
       promptBlockReason
     ) ? promptBlockReason : void 0;
     const isPromptBlocked = (candidate == null ? void 0 : candidate.finishReason) == null && confirmedPromptBlockReason != null;
-    const rawFinishReason = (_c = candidate == null ? void 0 : candidate.finishReason) != null ? _c : confirmedPromptBlockReason;
+    const rawFinishReason = (_h = candidate == null ? void 0 : candidate.finishReason) != null ? _h : confirmedPromptBlockReason;
     const content = [];
-    const parts = (_e = (_d = candidate == null ? void 0 : candidate.content) == null ? void 0 : _d.parts) != null ? _e : [];
-    const usageMetadata = response.usageMetadata;
+    const parts = [...continuedParts, ...(_j = (_i = candidate == null ? void 0 : candidate.content) == null ? void 0 : _i.parts) != null ? _j : []];
+    const usageMetadata = addGoogleGenerativeAIUsageMetadata(
+      continuedUsageMetadata,
+      (_k = response.usageMetadata) != null ? _k : void 0
+    );
     let lastCodeExecutionToolCallId;
     let lastServerToolCallId;
     for (const part of parts) {
-      if ("executableCode" in part && ((_f = part.executableCode) == null ? void 0 : _f.code)) {
+      if ("executableCode" in part && ((_l = part.executableCode) == null ? void 0 : _l.code)) {
         const toolCallId = this.config.generateId();
         lastCodeExecutionToolCallId = toolCallId;
         content.push({
@@ -2228,7 +2372,7 @@ var GoogleGenerativeAILanguageModel = class {
           toolName: "code_execution",
           result: {
             outcome: part.codeExecutionResult.outcome,
-            output: (_g = part.codeExecutionResult.output) != null ? _g : ""
+            output: (_m = part.codeExecutionResult.output) != null ? _m : ""
           }
         });
       } else if ("text" in part && part.text != null) {
@@ -2252,9 +2396,9 @@ var GoogleGenerativeAILanguageModel = class {
       } else if ("functionCall" in part && part.functionCall.name != null) {
         content.push({
           type: "tool-call",
-          toolCallId: (_h = part.functionCall.id) != null ? _h : this.config.generateId(),
+          toolCallId: (_n = part.functionCall.id) != null ? _n : this.config.generateId(),
           toolName: part.functionCall.name,
-          input: JSON.stringify((_i = part.functionCall.args) != null ? _i : {}),
+          input: JSON.stringify((_o = part.functionCall.args) != null ? _o : {}),
           providerMetadata: part.thoughtSignature ? {
             [providerOptionsName]: {
               thoughtSignature: part.thoughtSignature
@@ -2276,13 +2420,13 @@ var GoogleGenerativeAILanguageModel = class {
           } : void 0
         });
       } else if ("toolCall" in part && part.toolCall) {
-        const toolCallId = (_j = part.toolCall.id) != null ? _j : this.config.generateId();
+        const toolCallId = (_p = part.toolCall.id) != null ? _p : this.config.generateId();
         lastServerToolCallId = toolCallId;
         content.push({
           type: "tool-call",
           toolCallId,
           toolName: `server:${part.toolCall.toolType}`,
-          input: JSON.stringify((_k = part.toolCall.args) != null ? _k : {}),
+          input: JSON.stringify((_q = part.toolCall.args) != null ? _q : {}),
           providerExecuted: true,
           dynamic: true,
           providerMetadata: part.thoughtSignature ? {
@@ -2299,12 +2443,12 @@ var GoogleGenerativeAILanguageModel = class {
           }
         });
       } else if ("toolResponse" in part && part.toolResponse) {
-        const responseToolCallId = (_l = lastServerToolCallId != null ? lastServerToolCallId : part.toolResponse.id) != null ? _l : this.config.generateId();
+        const responseToolCallId = (_r = lastServerToolCallId != null ? lastServerToolCallId : part.toolResponse.id) != null ? _r : this.config.generateId();
         content.push({
           type: "tool-result",
           toolCallId: responseToolCallId,
           toolName: `server:${part.toolResponse.toolType}`,
-          result: (_m = part.toolResponse.response) != null ? _m : {},
+          result: (_s = part.toolResponse.response) != null ? _s : {},
           providerMetadata: part.thoughtSignature ? {
             [providerOptionsName]: {
               thoughtSignature: part.thoughtSignature,
@@ -2321,10 +2465,10 @@ var GoogleGenerativeAILanguageModel = class {
         lastServerToolCallId = void 0;
       }
     }
-    const sources = (_n = extractSources({
+    const sources = (_t = extractSources({
       groundingMetadata: candidate == null ? void 0 : candidate.groundingMetadata,
       generateId: this.config.generateId
-    })) != null ? _n : [];
+    })) != null ? _t : [];
     for (const source of sources) {
       content.push(source);
     }
@@ -2344,19 +2488,21 @@ var GoogleGenerativeAILanguageModel = class {
       warnings,
       providerMetadata: {
         [providerOptionsName]: {
-          promptFeedback: (_o = response.promptFeedback) != null ? _o : null,
-          groundingMetadata: (_p = candidate == null ? void 0 : candidate.groundingMetadata) != null ? _p : null,
-          urlContextMetadata: (_q = candidate == null ? void 0 : candidate.urlContextMetadata) != null ? _q : null,
-          safetyRatings: (_r = candidate == null ? void 0 : candidate.safetyRatings) != null ? _r : null,
+          promptFeedback: (_u = response.promptFeedback) != null ? _u : null,
+          groundingMetadata: (_v = candidate == null ? void 0 : candidate.groundingMetadata) != null ? _v : null,
+          urlContextMetadata: (_w = candidate == null ? void 0 : candidate.urlContextMetadata) != null ? _w : null,
+          safetyRatings: (_x = candidate == null ? void 0 : candidate.safetyRatings) != null ? _x : null,
           usageMetadata: usageMetadata != null ? usageMetadata : null,
-          finishMessage: (_s = candidate == null ? void 0 : candidate.finishMessage) != null ? _s : null,
-          serviceTier: (_t = usageMetadata == null ? void 0 : usageMetadata.serviceTier) != null ? _t : null
+          finishMessage: (_y = candidate == null ? void 0 : candidate.finishMessage) != null ? _y : null,
+          serviceTier: (_z = usageMetadata == null ? void 0 : usageMetadata.serviceTier) != null ? _z : null,
+          ...(candidate == null ? void 0 : candidate.continuationToken) != null ? { continuationToken: candidate.continuationToken } : {},
+          ...continuedUsageMetadata != null && firstPromptTokenCount != null ? { initialPromptTokenCount: firstPromptTokenCount } : {}
         }
       },
       request: { body: args },
       response: {
         // TODO timestamp, model id
-        id: (_u = response.responseId) != null ? _u : void 0,
+        id: (_A = response.responseId) != null ? _A : void 0,
         headers: responseHeaders,
         body: rawResponse
       }
@@ -2369,27 +2515,40 @@ var GoogleGenerativeAILanguageModel = class {
       options.headers,
       extraHeaders
     );
-    const { responseHeaders, value: response } = await (0, import_provider_utils8.postJsonToApi)({
+    const streamGenerateContent = (body) => (0, import_provider_utils8.postJsonToApi)({
       url: `${this.config.baseURL}/${getModelPath(
         this.modelId
       )}:streamGenerateContent?alt=sse`,
       headers,
-      body: args,
+      body,
       failedResponseHandler: googleFailedResponseHandler,
       successfulResponseHandler: (0, import_provider_utils8.createEventSourceResponseHandler)(chunkSchema),
       abortSignal: options.abortSignal,
       fetch: this.config.fetch
     });
+    const { responseHeaders, value: firstResponse } = await streamGenerateContent(args);
+    const response = continueGoogleStream({
+      stream: firstResponse,
+      getContinuationToken: (chunk) => {
+        var _a2, _b;
+        const candidate = (_a2 = chunk.candidates) == null ? void 0 : _a2[0];
+        return (candidate == null ? void 0 : candidate.finishReason) === "CONTINUATION" ? (_b = candidate.continuationToken) != null ? _b : void 0 : void 0;
+      },
+      continueStream: async (continuationToken) => (await streamGenerateContent({ ...args, continuationToken })).value
+    });
     let finishReason = {
       unified: "other",
       raw: void 0
     };
     let usage = void 0;
+    let continuedUsage = void 0;
+    let initialPromptTokenCount = void 0;
     let promptFeedback = null;
     let lastGroundingMetadata = null;
     let lastUrlContextMetadata = null;
     let lastSafetyRatings = null;
     let lastFinishMessage = null;
+    let lastContinuationToken = null;
     let confirmedPromptBlockReason;
     const generateId3 = this.config.generateId;
     let hasToolCalls = false;
@@ -2480,6 +2639,9 @@ var GoogleGenerativeAILanguageModel = class {
               if (candidate.finishMessage != null) {
                 lastFinishMessage = candidate.finishMessage;
               }
+              if (candidate.continuationToken != null) {
+                lastContinuationToken = candidate.continuationToken;
+              }
             }
             if (confirmedPromptBlockReason != null || candidate == null) {
               return;
@@ -2764,6 +2926,16 @@ var GoogleGenerativeAILanguageModel = class {
                 }
               }
             }
+            if (candidate.finishReason === "CONTINUATION") {
+              if (continuedUsage == null) {
+                initialPromptTokenCount = usage == null ? void 0 : usage.promptTokenCount;
+              }
+              continuedUsage = addGoogleGenerativeAIUsageMetadata(
+                continuedUsage,
+                usage
+              );
+              usage = void 0;
+            }
             if (candidate.finishReason != null) {
               finishReason = {
                 unified: mapGoogleGenerativeAIFinishReason({
@@ -2776,6 +2948,7 @@ var GoogleGenerativeAILanguageModel = class {
           },
           flush(controller) {
             var _a2;
+            usage = addGoogleGenerativeAIUsageMetadata(continuedUsage, usage);
             if (currentTextBlockId !== null) {
               controller.enqueue({
                 type: "text-end",
@@ -2800,7 +2973,12 @@ var GoogleGenerativeAILanguageModel = class {
                   safetyRatings: lastSafetyRatings,
                   usageMetadata: usage != null ? usage : null,
                   finishMessage: lastFinishMessage,
-                  serviceTier: (_a2 = usage == null ? void 0 : usage.serviceTier) != null ? _a2 : null
+                  serviceTier: (_a2 = usage == null ? void 0 : usage.serviceTier) != null ? _a2 : null,
+                  // Only surface the token when the response stopped early
+                  // (budget used up, or continuing failed), not for the
+                  // checkpoint tokens of a stream that finished normally.
+                  ...lastContinuationToken != null && (finishReason.raw === "MAX_TOKENS" || finishReason.raw === "CONTINUATION") ? { continuationToken: lastContinuationToken } : {},
+                  ...initialPromptTokenCount != null ? { initialPromptTokenCount } : {}
                 }
               }
             });
@@ -2924,7 +3102,15 @@ var getGroundingMetadataSchema = () => import_v47.z.object({
         uri: import_v47.z.string().nullish(),
         title: import_v47.z.string().nullish(),
         text: import_v47.z.string().nullish(),
-        fileSearchStore: import_v47.z.string().nullish()
+        fileSearchStore: import_v47.z.string().nullish(),
+        customMetadata: import_v47.z.array(
+          import_v47.z.object({
+            key: import_v47.z.string(),
+            stringValue: import_v47.z.string().nullish(),
+            numericValue: import_v47.z.number().nullish(),
+            stringListValue: import_v47.z.object({ values: import_v47.z.array(import_v47.z.string()).nullish() }).nullish()
+          })
+        ).nullish()
       }).nullish(),
       maps: import_v47.z.object({
         uri: import_v47.z.string().nullish(),
@@ -3064,6 +3250,7 @@ var responseSchema = (0, import_provider_utils8.lazySchema)(
           content: getContentSchema().nullish().or(import_v47.z.object({}).strict()),
           finishReason: import_v47.z.string().nullish(),
           finishMessage: import_v47.z.string().nullish(),
+          continuationToken: import_v47.z.string().nullish(),
           safetyRatings: import_v47.z.array(getSafetyRatingSchema()).nullish(),
           groundingMetadata: getGroundingMetadataSchema().nullish(),
           urlContextMetadata: getUrlContextMetadataSchema().nullish()
@@ -3086,6 +3273,7 @@ var chunkSchema = (0, import_provider_utils8.lazySchema)(
           content: getContentSchema().nullish(),
           finishReason: import_v47.z.string().nullish(),
           finishMessage: import_v47.z.string().nullish(),
+          continuationToken: import_v47.z.string().nullish(),
           safetyRatings: import_v47.z.array(getSafetyRatingSchema()).nullish(),
           groundingMetadata: getGroundingMetadataSchema().nullish(),
           urlContextMetadata: getUrlContextMetadataSchema().nullish()
@@ -6567,7 +6755,7 @@ var GoogleInteractionsLanguageModel = class {
   }
   async doGenerate(options) {
     var _a2, _b, _c, _d, _e, _f;
-    const { args, warnings, isAgent, pollingTimeoutMs } = await this.getArgs(options);
+    const { args, warnings, isAgent, isBackground, pollingTimeoutMs } = await this.getArgs(options);
     const url = `${this.config.baseURL}/interactions`;
     const mergedHeaders = (0, import_provider_utils24.combineHeaders)(
       INTERACTIONS_API_REVISION_HEADER,
@@ -6590,7 +6778,7 @@ var GoogleInteractionsLanguageModel = class {
       value: response,
       rawValue: rawResponse
     } = postResult;
-    if (isAgent && !isTerminalStatus(response.status)) {
+    if ((isAgent || isBackground) && !isTerminalStatus(response.status)) {
       const polled = await pollGoogleInteractionUntilTerminal({
         baseURL: this.config.baseURL,
         interactionId: response.id,