@ai-sdk/google 4.0.93 → 4.0.95

package/src/google-model-capabilities.ts

@@ -4,16 +4,16 @@ type GoogleModelCapabilities = {
   usesGemini3Features: boolean;
 };
 
-const gemini1ModelPattern = /(^|\/)gemini-1(?:[.-]|$)/i;
-const gemini2ModelPattern = /(^|\/)gemini-2(?:[.-]|$)/i;
-const gemini25ModelPattern = /(^|\/)gemini-2\.5(?:[.-]|$)/i;
-const geminiModelPattern = /(^|\/)gemini-/i;
+const gemini1ModelPattern = /(^|[/.])gemini-1(?:[.-]|$)/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)
+    /(^|[/.])gemini-pro(?:-vision)?$/i.test(modelId) ||
+    /(^|[/.])gemini-robotics-er-1\.5(?:[.-]|$)/i.test(modelId)
   );
 }
 

package/src/google-prompt.ts

@@ -100,4 +100,22 @@ export interface GoogleProviderMetadata {
   finishReason?: string | 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;
 }

package/src/interactions/google-interactions-language-model-options.ts

@@ -37,6 +37,7 @@ export type GoogleInteractionsModelId =
   | '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'

package/src/interactions/google-interactions-language-model.ts

@@ -482,7 +482,7 @@ export class GoogleInteractionsLanguageModel implements LanguageModelV4 {
   async doGenerate(
     options: LanguageModelV4CallOptions,
   ): Promise<LanguageModelV4GenerateResult> {
-    const { args, warnings, isAgent, pollingTimeoutMs } =
+    const { args, warnings, isAgent, isBackground, pollingTimeoutMs } =
       await this.getArgs(options);
 
     const url = `${this.config.baseURL}/interactions`;
@@ -511,12 +511,12 @@ export class GoogleInteractionsLanguageModel implements LanguageModelV4 {
     } = postResult;
 
     /*
-     * Agent calls may return a non-terminal status (`in_progress` /
-     * `requires_action`) when invoked with `background: true`. Poll
-     * `GET /interactions/{id}` until terminal so the user-facing surface
-     * matches a synchronous call.
+     * Agent calls and `background: true` calls (e.g. Gemini 4 long decode,
+     * where the server runs every hop) may return a non-terminal status
+     * (`in_progress` / `requires_action`). Poll `GET /interactions/{id}` until
+     * terminal so the user-facing surface matches a synchronous call.
      */
-    if (isAgent && !isTerminalStatus(response.status)) {
+    if ((isAgent || isBackground) && !isTerminalStatus(response.status)) {
       const polled = await pollGoogleInteractionUntilTerminal({
         baseURL: this.config.baseURL,
         interactionId: response.id,

package/src/map-google-finish-reason.ts

@@ -11,6 +11,9 @@ export function mapGoogleFinishReason({
     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':