@ai-sdk/google 4.0.0-canary.69 → 4.0.0-canary.71

package/CHANGELOG.md

@@ -1,5 +1,23 @@
 # @ai-sdk/google
 
+## 4.0.0-canary.71
+
+### Patch Changes
+
+- 045d2e8: fix(google): read `serviceTier` from `usageMetadata.serviceTier` in both generate and stream paths
+
+  The previous implementation read `serviceTier` from the `x-gemini-service-tier`
+  response header, which is only populated on non-streaming responses. Gemini
+  streaming includes the value in `usageMetadata.serviceTier` on every chunk, so
+  `providerMetadata.google.serviceTier` was always `null` for streams. Read from
+  `usageMetadata` for both paths instead.
+
+## 4.0.0-canary.70
+
+### Patch Changes
+
+- aeea161: feat(google): read `serviceTier` from `x-gemini-service-tier` response header in Gemini API and use PayGo for Vertex
+
 ## 4.0.0-canary.69
 
 ### Patch Changes

package/dist/index.d.ts

@@ -41,6 +41,8 @@ declare const googleLanguageModelOptions: _ai_sdk_provider_utils.LazySchema<{
     } | undefined;
     streamFunctionCallArguments?: boolean | undefined;
     serviceTier?: "standard" | "flex" | "priority" | undefined;
+    sharedRequestType?: "standard" | "flex" | "priority" | undefined;
+    requestType?: "shared" | undefined;
 }>;
 type GoogleLanguageModelOptions = InferSchema<typeof googleLanguageModelOptions>;
 
@@ -169,6 +171,7 @@ declare const responseSchema: _ai_sdk_provider_utils.LazySchema<{
         candidatesTokenCount?: number | null | undefined;
         totalTokenCount?: number | null | undefined;
         trafficType?: string | null | undefined;
+        serviceTier?: string | null | undefined;
         promptTokensDetails?: {
             modality: string;
             tokenCount: number;
@@ -189,7 +192,6 @@ declare const responseSchema: _ai_sdk_provider_utils.LazySchema<{
             blocked?: boolean | null | undefined;
         }[] | null | undefined;
     } | null | undefined;
-    serviceTier?: string | null | undefined;
 }>;
 type GroundingMetadataSchema = NonNullable<InferSchema<typeof responseSchema>['candidates'][number]['groundingMetadata']>;
 type UrlContextMetadataSchema = NonNullable<InferSchema<typeof responseSchema>['candidates'][number]['urlContextMetadata']>;

package/dist/index.js

@@ -7,7 +7,7 @@ import {
 } from "@ai-sdk/provider-utils";
 
 // src/version.ts
-var VERSION = true ? "4.0.0-canary.69" : "0.0.0-test";
+var VERSION = true ? "4.0.0-canary.71" : "0.0.0-test";
 
 // src/google-embedding-model.ts
 import {
@@ -989,17 +989,32 @@ var googleLanguageModelOptions = lazySchema4(
        */
       streamFunctionCallArguments: z4.boolean().optional(),
       /**
-       * Optional. The service tier to use for the request.
+       * Optional. The service tier to use for the request. Sent as the
+       * `serviceTier` body field. Gemini API only.
        */
-      serviceTier: z4.enum(["standard", "flex", "priority"]).optional()
+      serviceTier: z4.enum(["standard", "flex", "priority"]).optional(),
+      /**
+       * Optional. Vertex AI only. Sent as the
+       * `X-Vertex-AI-LLM-Shared-Request-Type` request header to select a
+       * shared (PayGo) tier. With Provisioned Throughput allocated and
+       * `requestType` unset, the request falls back to this tier only if
+       * PT capacity is exhausted.
+       *
+       * https://docs.cloud.google.com/vertex-ai/generative-ai/docs/priority-paygo
+       * https://docs.cloud.google.com/vertex-ai/generative-ai/docs/flex-paygo
+       */
+      sharedRequestType: z4.enum(["priority", "flex", "standard"]).optional(),
+      /**
+       * Optional. Vertex AI only. Sent as the `X-Vertex-AI-LLM-Request-Type`
+       * request header. Set to `'shared'` together with `sharedRequestType`
+       * to bypass Provisioned Throughput entirely.
+       *
+       * https://docs.cloud.google.com/vertex-ai/generative-ai/docs/priority-paygo
+       */
+      requestType: z4.enum(["shared"]).optional()
     })
   )
 );
-var VertexServiceTierMap = {
-  standard: "SERVICE_TIER_STANDARD",
-  flex: "SERVICE_TIER_FLEX",
-  priority: "SERVICE_TIER_PRIORITY"
-};
 
 // src/google-prepare-tools.ts
 import {
@@ -1580,10 +1595,27 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
         message: `'streamFunctionCallArguments' is only supported on the Vertex AI API and will be ignored with the current Google provider (${this.config.provider}). See https://docs.cloud.google.com/vertex-ai/generative-ai/docs/multimodal/function-calling#streaming-fc`
       });
     }
-    let sanitizedServiceTier = googleOptions == null ? void 0 : googleOptions.serviceTier;
     if ((googleOptions == null ? void 0 : googleOptions.serviceTier) && isVertexProvider) {
-      sanitizedServiceTier = VertexServiceTierMap[googleOptions.serviceTier];
+      warnings.push({
+        type: "other",
+        message: "'serviceTier' is a Gemini API option and is not supported on Vertex AI. Use 'sharedRequestType' (and optionally 'requestType') instead. See https://docs.cloud.google.com/vertex-ai/generative-ai/docs/priority-paygo"
+      });
+    }
+    if (((googleOptions == null ? void 0 : googleOptions.sharedRequestType) || (googleOptions == null ? void 0 : googleOptions.requestType)) && !isVertexProvider) {
+      warnings.push({
+        type: "other",
+        message: `'sharedRequestType' and 'requestType' are Vertex AI options and are ignored with the current Google provider (${this.config.provider}).`
+      });
     }
+    const vertexPaygoHeaders = isVertexProvider && ((googleOptions == null ? void 0 : googleOptions.sharedRequestType) || (googleOptions == null ? void 0 : googleOptions.requestType)) ? {
+      ...googleOptions.sharedRequestType && {
+        "X-Vertex-AI-LLM-Shared-Request-Type": googleOptions.sharedRequestType
+      },
+      ...googleOptions.requestType && {
+        "X-Vertex-AI-LLM-Request-Type": googleOptions.requestType
+      }
+    } : void 0;
+    const bodyServiceTier = isVertexProvider ? void 0 : googleOptions == null ? void 0 : googleOptions.serviceTier;
     const isGemmaModel = this.modelId.toLowerCase().startsWith("gemma-");
     const supportsFunctionResponseParts = this.modelId.startsWith("gemini-3");
     const { contents, systemInstruction } = convertToGoogleMessages(prompt, {
@@ -1658,21 +1690,23 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
         toolConfig,
         cachedContent: googleOptions == null ? void 0 : googleOptions.cachedContent,
         labels: googleOptions == null ? void 0 : googleOptions.labels,
-        serviceTier: sanitizedServiceTier
+        serviceTier: bodyServiceTier
       },
       warnings: [...warnings, ...toolWarnings],
-      providerOptionsNames
+      providerOptionsNames,
+      extraHeaders: vertexPaygoHeaders
     };
   }
   async doGenerate(options) {
     var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r;
-    const { args, warnings, providerOptionsNames } = await this.getArgs(options);
+    const { args, warnings, providerOptionsNames, extraHeaders } = await this.getArgs(options);
     const wrapProviderMetadata = (payload) => Object.fromEntries(
       providerOptionsNames.map((name) => [name, payload])
     );
     const mergedHeaders = combineHeaders2(
       this.config.headers ? await resolve2(this.config.headers) : void 0,
-      options.headers
+      options.headers,
+      extraHeaders
     );
     const {
       responseHeaders,
@@ -1821,7 +1855,7 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
         safetyRatings: (_p = candidate.safetyRatings) != null ? _p : null,
         usageMetadata: usageMetadata != null ? usageMetadata : null,
         finishMessage: (_q = candidate.finishMessage) != null ? _q : null,
-        serviceTier: (_r = response.serviceTier) != null ? _r : null
+        serviceTier: (_r = usageMetadata == null ? void 0 : usageMetadata.serviceTier) != null ? _r : null
       }),
       request: { body: args },
       response: {
@@ -1832,16 +1866,14 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
     };
   }
   async doStream(options) {
-    const { args, warnings, providerOptionsNames } = await this.getArgs(
-      options,
-      { isStreaming: true }
-    );
+    const { args, warnings, providerOptionsNames, extraHeaders } = await this.getArgs(options, { isStreaming: true });
     const wrapProviderMetadata = (payload) => Object.fromEntries(
       providerOptionsNames.map((name) => [name, payload])
     );
     const headers = combineHeaders2(
       this.config.headers ? await resolve2(this.config.headers) : void 0,
-      options.headers
+      options.headers,
+      extraHeaders
     );
     const { responseHeaders, value: response } = await postJsonToApi2({
       url: `${this.config.baseURL}/${getModelPath(
@@ -1862,7 +1894,6 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
     let providerMetadata = void 0;
     let lastGroundingMetadata = null;
     let lastUrlContextMetadata = null;
-    let serviceTier = null;
     const generateId3 = this.config.generateId;
     let hasToolCalls = false;
     let currentTextBlockId = null;
@@ -1907,7 +1938,7 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
             controller.enqueue({ type: "stream-start", warnings });
           },
           transform(chunk, controller) {
-            var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o;
+            var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p;
             if (options.includeRawChunks) {
               controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
             }
@@ -1920,9 +1951,6 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
             if (usageMetadata != null) {
               usage = usageMetadata;
             }
-            if (value.serviceTier != null) {
-              serviceTier = value.serviceTier;
-            }
             const candidate = (_a = value.candidates) == null ? void 0 : _a[0];
             if (candidate == null) {
               return;
@@ -2218,7 +2246,7 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
                 safetyRatings: (_n = candidate.safetyRatings) != null ? _n : null,
                 usageMetadata: usageMetadata != null ? usageMetadata : null,
                 finishMessage: (_o = candidate.finishMessage) != null ? _o : null,
-                serviceTier
+                serviceTier: (_p = usage == null ? void 0 : usage.serviceTier) != null ? _p : null
               });
             }
           },
@@ -2545,6 +2573,7 @@ var usageSchema = z5.object({
   totalTokenCount: z5.number().nullish(),
   // https://cloud.google.com/vertex-ai/generative-ai/docs/reference/rest/v1/GenerateContentResponse#TrafficType
   trafficType: z5.string().nullish(),
+  serviceTier: z5.string().nullish(),
   // https://ai.google.dev/api/generate-content#Modality
   promptTokensDetails: tokenDetailsSchema,
   candidatesTokensDetails: tokenDetailsSchema
@@ -2574,8 +2603,7 @@ var responseSchema = lazySchema5(
       promptFeedback: z5.object({
         blockReason: z5.string().nullish(),
         safetyRatings: z5.array(getSafetyRatingSchema()).nullish()
-      }).nullish(),
-      serviceTier: z5.string().nullish()
+      }).nullish()
     })
   )
 );
@@ -2596,8 +2624,7 @@ var chunkSchema = lazySchema5(
       promptFeedback: z5.object({
         blockReason: z5.string().nullish(),
         safetyRatings: z5.array(getSafetyRatingSchema()).nullish()
-      }).nullish(),
-      serviceTier: z5.string().nullish()
+      }).nullish()
     })
   )
 );