@ai-sdk/gateway 4.0.0-beta.5 → 4.0.0-beta.51

package/src/gateway-model-entry.ts

@@ -1,4 +1,4 @@
-import type { LanguageModelV3 } from '@ai-sdk/provider';
+import type { LanguageModelV4 } from '@ai-sdk/provider';
 
 export interface GatewayLanguageModelEntry {
   /**
@@ -53,6 +53,6 @@ export interface GatewayLanguageModelEntry {
 }
 
 export type GatewayLanguageModelSpecification = Pick<
-  LanguageModelV3,
+  LanguageModelV4,
   'specificationVersion' | 'provider' | 'modelId'
 >;

package/src/gateway-provider-options.ts

@@ -2,7 +2,7 @@ import { InferSchema, lazySchema, zodSchema } from '@ai-sdk/provider-utils';
 import { z } from 'zod/v4';
 
 // https://vercel.com/docs/ai-gateway/provider-options
-const gatewayLanguageModelOptions = lazySchema(() =>
+const gatewayProviderOptions = lazySchema(() =>
   zodSchema(
     z.object({
       /**
@@ -17,6 +17,14 @@ const gatewayLanguageModelOptions = lazySchema(() =>
        * Example: `['bedrock', 'anthropic']` will try Amazon Bedrock first, then Anthropic as fallback.
        */
       order: z.array(z.string()).optional(),
+      /**
+       * Sort providers by a performance or cost metric before routing.
+       *
+       * - `'cost'`: lowest cost first
+       * - `'ttft'`: lowest time-to-first-token first
+       * - `'tps'`: highest tokens-per-second first
+       */
+      sort: z.enum(['cost', 'ttft', 'tps']).optional(),
       /**
        * The unique identifier for the end user on behalf of whom the request was made.
        *
@@ -53,12 +61,33 @@ const gatewayLanguageModelOptions = lazySchema(() =>
         .record(z.string(), z.array(z.record(z.string(), z.unknown())))
         .optional(),
       /**
-       * Whether to filter by only providers that state they have zero data
-       * retention with Vercel AI Gateway. When enabled, only providers that
-       * have agreements with Vercel AI Gateway for zero data retention will be
-       * used.
+       * Whether to filter by only providers that have zero data retention
+       * agreements with Vercel for AI Gateway. When using BYOK credentials,
+       * this filter is not applied. If BYOK credentials fail and the request
+       * falls back to system credentials, only providers with zero data
+       * retention agreements will be used.
        */
       zeroDataRetention: z.boolean().optional(),
+      /**
+       * Whether to filter by only providers that do not train on prompt data.
+       * When using BYOK credentials, this filter is not applied. If BYOK
+       * credentials fail and the request falls back to system credentials,
+       * only providers that have agreements with Vercel for AI Gateway to not
+       * use prompts for model training will be used.
+       */
+      disallowPromptTraining: z.boolean().optional(),
+      /**
+       * Whether to filter by only providers that are HIPAA compliant with
+       * Vercel AI Gateway. When enabled, only providers that have agreements
+       * with Vercel AI Gateway for HIPAA compliance will be used.
+       */
+      hipaaCompliant: z.boolean().optional(),
+      /**
+       * The unique identifier for the entity against which quota is tracked.
+       *
+       * Used for quota management and enforcement purposes.
+       */
+      quotaEntityId: z.string().optional(),
       /**
        * Per-provider timeouts for BYOK credentials in milliseconds.
        * Controls how long to wait for a provider to start responding
@@ -75,6 +104,4 @@ const gatewayLanguageModelOptions = lazySchema(() =>
   ),
 );
 
-export type GatewayLanguageModelOptions = InferSchema<
-  typeof gatewayLanguageModelOptions
->;
+export type GatewayProviderOptions = InferSchema<typeof gatewayProviderOptions>;

package/src/gateway-provider.ts

@@ -13,38 +13,51 @@ import {
   type GatewayFetchMetadataResponse,
   type GatewayCreditsResponse,
 } from './gateway-fetch-metadata';
+import {
+  GatewaySpendReport,
+  type GatewaySpendReportParams,
+  type GatewaySpendReportResponse,
+} from './gateway-spend-report';
+import {
+  GatewayGenerationInfoFetcher,
+  type GatewayGenerationInfoParams,
+  type GatewayGenerationInfo,
+} from './gateway-generation-info';
 import { GatewayLanguageModel } from './gateway-language-model';
 import { GatewayEmbeddingModel } from './gateway-embedding-model';
 import { GatewayImageModel } from './gateway-image-model';
 import { GatewayVideoModel } from './gateway-video-model';
+import { GatewayRerankingModel } from './gateway-reranking-model';
 import type { GatewayEmbeddingModelId } from './gateway-embedding-model-settings';
 import type { GatewayImageModelId } from './gateway-image-model-settings';
+import type { GatewayRerankingModelId } from './gateway-reranking-model-settings';
 import type { GatewayVideoModelId } from './gateway-video-model-settings';
 import { gatewayTools } from './gateway-tools';
 import { getVercelOidcToken, getVercelRequestId } from './vercel-environment';
 import type { GatewayModelId } from './gateway-language-model-settings';
 import type {
-  LanguageModelV3,
-  EmbeddingModelV3,
-  ImageModelV3,
-  Experimental_VideoModelV3,
-  ProviderV3,
+  LanguageModelV4,
+  EmbeddingModelV4,
+  ImageModelV4,
+  RerankingModelV4,
+  Experimental_VideoModelV4,
+  ProviderV4,
 } from '@ai-sdk/provider';
 import { withUserAgentSuffix } from '@ai-sdk/provider-utils';
 import { VERSION } from './version';
 
-export interface GatewayProvider extends ProviderV3 {
-  (modelId: GatewayModelId): LanguageModelV3;
+export interface GatewayProvider extends ProviderV4 {
+  (modelId: GatewayModelId): LanguageModelV4;
 
   /**
    * Creates a model for text generation.
    */
-  chat(modelId: GatewayModelId): LanguageModelV3;
+  chat(modelId: GatewayModelId): LanguageModelV4;
 
   /**
    * Creates a model for text generation.
    */
-  languageModel(modelId: GatewayModelId): LanguageModelV3;
+  languageModel(modelId: GatewayModelId): LanguageModelV4;
 
   /**
    * Returns available providers and models for use with the remote provider.
@@ -56,40 +69,66 @@ export interface GatewayProvider extends ProviderV3 {
    */
   getCredits(): Promise<GatewayCreditsResponse>;
 
+  /**
+   * Returns a spend report with cost, token, and request count data,
+   * aggregated by the specified dimension.
+   */
+  getSpendReport(
+    params: GatewaySpendReportParams,
+  ): Promise<GatewaySpendReportResponse>;
+
+  /**
+   * Returns detailed information about a specific generation by its ID,
+   * including cost, token usage, latency, and provider details.
+   */
+  getGenerationInfo(
+    params: GatewayGenerationInfoParams,
+  ): Promise<GatewayGenerationInfo>;
+
   /**
    * Creates a model for generating text embeddings.
    */
-  embedding(modelId: GatewayEmbeddingModelId): EmbeddingModelV3;
+  embedding(modelId: GatewayEmbeddingModelId): EmbeddingModelV4;
 
   /**
    * Creates a model for generating text embeddings.
    */
-  embeddingModel(modelId: GatewayEmbeddingModelId): EmbeddingModelV3;
+  embeddingModel(modelId: GatewayEmbeddingModelId): EmbeddingModelV4;
 
   /**
    * @deprecated Use `embeddingModel` instead.
    */
-  textEmbeddingModel(modelId: GatewayEmbeddingModelId): EmbeddingModelV3;
+  textEmbeddingModel(modelId: GatewayEmbeddingModelId): EmbeddingModelV4;
 
   /**
    * Creates a model for generating images.
    */
-  image(modelId: GatewayImageModelId): ImageModelV3;
+  image(modelId: GatewayImageModelId): ImageModelV4;
 
   /**
    * Creates a model for generating images.
    */
-  imageModel(modelId: GatewayImageModelId): ImageModelV3;
+  imageModel(modelId: GatewayImageModelId): ImageModelV4;
 
   /**
    * Creates a model for generating videos.
    */
-  video(modelId: GatewayVideoModelId): Experimental_VideoModelV3;
+  video(modelId: GatewayVideoModelId): Experimental_VideoModelV4;
 
   /**
    * Creates a model for generating videos.
    */
-  videoModel(modelId: GatewayVideoModelId): Experimental_VideoModelV3;
+  videoModel(modelId: GatewayVideoModelId): Experimental_VideoModelV4;
+
+  /**
+   * Creates a model for reranking documents.
+   */
+  reranking(modelId: GatewayRerankingModelId): RerankingModelV4;
+
+  /**
+   * Creates a model for reranking documents.
+   */
+  rerankingModel(modelId: GatewayRerankingModelId): RerankingModelV4;
 
   /**
    * Gateway-specific tools executed server-side.
@@ -253,6 +292,36 @@ export function createGatewayProvider(
       });
   };
 
+  const getSpendReport = async (params: GatewaySpendReportParams) => {
+    return new GatewaySpendReport({
+      baseURL,
+      headers: getHeaders,
+      fetch: options.fetch,
+    })
+      .getSpendReport(params)
+      .catch(async (error: unknown) => {
+        throw await asGatewayError(
+          error,
+          await parseAuthMethod(await getHeaders()),
+        );
+      });
+  };
+
+  const getGenerationInfo = async (params: GatewayGenerationInfoParams) => {
+    return new GatewayGenerationInfoFetcher({
+      baseURL,
+      headers: getHeaders,
+      fetch: options.fetch,
+    })
+      .getGenerationInfo(params)
+      .catch(async (error: unknown) => {
+        throw await asGatewayError(
+          error,
+          await parseAuthMethod(await getHeaders()),
+        );
+      });
+  };
+
   const provider = function (modelId: GatewayModelId) {
     if (new.target) {
       throw new Error(
@@ -263,9 +332,11 @@ export function createGatewayProvider(
     return createLanguageModel(modelId);
   };
 
-  provider.specificationVersion = 'v3' as const;
+  provider.specificationVersion = 'v4' as const;
   provider.getAvailableModels = getAvailableModels;
   provider.getCredits = getCredits;
+  provider.getSpendReport = getSpendReport;
+  provider.getGenerationInfo = getGenerationInfo;
   provider.imageModel = (modelId: GatewayImageModelId) => {
     return new GatewayImageModel(modelId, {
       provider: 'gateway',
@@ -296,6 +367,17 @@ export function createGatewayProvider(
       o11yHeaders: createO11yHeaders(),
     });
   };
+  const createRerankingModel = (modelId: GatewayRerankingModelId) => {
+    return new GatewayRerankingModel(modelId, {
+      provider: 'gateway',
+      baseURL,
+      headers: getHeaders,
+      fetch: options.fetch,
+      o11yHeaders: createO11yHeaders(),
+    });
+  };
+  provider.rerankingModel = createRerankingModel;
+  provider.reranking = createRerankingModel;
   provider.chat = provider.languageModel;
   provider.embedding = provider.embeddingModel;
   provider.image = provider.imageModel;

package/src/gateway-reranking-model-settings.ts

@@ -0,0 +1,7 @@
+export type GatewayRerankingModelId =
+  | 'cohere/rerank-v3.5'
+  | 'cohere/rerank-v4-fast'
+  | 'cohere/rerank-v4-pro'
+  | 'voyage/rerank-2.5'
+  | 'voyage/rerank-2.5-lite'
+  | (string & {});

package/src/gateway-reranking-model.ts

@@ -0,0 +1,119 @@
+import type {
+  RerankingModelV4,
+  SharedV4ProviderMetadata,
+} from '@ai-sdk/provider';
+import {
+  combineHeaders,
+  createJsonErrorResponseHandler,
+  createJsonResponseHandler,
+  lazySchema,
+  postJsonToApi,
+  resolve,
+  zodSchema,
+  type Resolvable,
+} from '@ai-sdk/provider-utils';
+import { z } from 'zod/v4';
+import { asGatewayError } from './errors';
+import { parseAuthMethod } from './errors/parse-auth-method';
+import type { GatewayConfig } from './gateway-config';
+
+export class GatewayRerankingModel implements RerankingModelV4 {
+  readonly specificationVersion = 'v4';
+
+  constructor(
+    readonly modelId: string,
+    private readonly config: GatewayConfig & {
+      provider: string;
+      o11yHeaders: Resolvable<Record<string, string>>;
+    },
+  ) {}
+
+  get provider(): string {
+    return this.config.provider;
+  }
+
+  async doRerank({
+    documents,
+    query,
+    topN,
+    headers,
+    abortSignal,
+    providerOptions,
+  }: Parameters<RerankingModelV4['doRerank']>[0]): Promise<
+    Awaited<ReturnType<RerankingModelV4['doRerank']>>
+  > {
+    const resolvedHeaders = this.config.headers
+      ? await resolve(this.config.headers)
+      : undefined;
+    try {
+      const {
+        responseHeaders,
+        value: responseBody,
+        rawValue,
+      } = await postJsonToApi({
+        url: this.getUrl(),
+        headers: combineHeaders(
+          resolvedHeaders,
+          headers ?? {},
+          this.getModelConfigHeaders(),
+          await resolve(this.config.o11yHeaders),
+        ),
+        body: {
+          documents,
+          query,
+          ...(topN != null ? { topN } : {}),
+          ...(providerOptions ? { providerOptions } : {}),
+        },
+        successfulResponseHandler: createJsonResponseHandler(
+          gatewayRerankingResponseSchema,
+        ),
+        failedResponseHandler: createJsonErrorResponseHandler({
+          errorSchema: z.any(),
+          errorToMessage: data => data,
+        }),
+        ...(abortSignal && { abortSignal }),
+        fetch: this.config.fetch,
+      });
+
+      return {
+        ranking: responseBody.ranking,
+        providerMetadata:
+          responseBody.providerMetadata as unknown as SharedV4ProviderMetadata,
+        response: { headers: responseHeaders, body: rawValue },
+        warnings: [],
+      };
+    } catch (error) {
+      throw await asGatewayError(
+        error,
+        await parseAuthMethod(resolvedHeaders ?? {}),
+      );
+    }
+  }
+
+  private getUrl() {
+    return `${this.config.baseURL}/reranking-model`;
+  }
+
+  private getModelConfigHeaders() {
+    return {
+      'ai-reranking-model-specification-version': '4',
+      'ai-model-id': this.modelId,
+    };
+  }
+}
+
+const gatewayRerankingResponseSchema = lazySchema(() =>
+  zodSchema(
+    z.object({
+      ranking: z.array(
+        z.object({
+          index: z.number(),
+          relevanceScore: z.number(),
+        }),
+      ),
+      providerMetadata: z
+        .record(z.string(), z.record(z.string(), z.unknown()))
+        .optional(),
+    }),
+  ),
+);

package/src/gateway-spend-report.ts

@@ -0,0 +1,193 @@
+import {
+  createJsonErrorResponseHandler,
+  createJsonResponseHandler,
+  getFromApi,
+  lazySchema,
+  resolve,
+  zodSchema,
+} from '@ai-sdk/provider-utils';
+import { z } from 'zod/v4';
+import { asGatewayError } from './errors';
+import type { GatewayConfig } from './gateway-config';
+
+export interface GatewaySpendReportParams {
+  /** Start date in YYYY-MM-DD format (inclusive) */
+  startDate: string;
+  /** End date in YYYY-MM-DD format (inclusive) */
+  endDate: string;
+  /** Primary aggregation dimension. Defaults to 'day'. */
+  groupBy?: 'day' | 'user' | 'model' | 'tag' | 'provider' | 'credential_type';
+  /** Time granularity when groupBy is 'day'. */
+  datePart?: 'day' | 'hour';
+  /** Filter to a specific user's spend. */
+  userId?: string;
+  /** Filter to a specific model (e.g. 'anthropic/claude-sonnet-4.5'). */
+  model?: string;
+  /** Filter to a specific provider (e.g. 'anthropic'). */
+  provider?: string;
+  /** Filter to BYOK or system credentials. */
+  credentialType?: 'byok' | 'system';
+  /** Filter to requests with these tags. */
+  tags?: string[];
+}
+
+export interface GatewaySpendReportRow {
+  /** Date string (present when groupBy is 'day') */
+  day?: string;
+  /** Hour timestamp (present when groupBy is 'day' and datePart is 'hour') */
+  hour?: string;
+  /** User identifier (present when groupBy is 'user') */
+  user?: string;
+  /** Model identifier (present when groupBy is 'model') */
+  model?: string;
+  /** Tag value (present when groupBy is 'tag') */
+  tag?: string;
+  /** Provider name (present when groupBy is 'provider') */
+  provider?: string;
+  /** Credential type (present when groupBy is 'credential_type') */
+  credentialType?: 'byok' | 'system';
+
+  /** Total cost in USD */
+  totalCost: number;
+  /** Market cost in USD */
+  marketCost?: number;
+  /** Number of input tokens */
+  inputTokens?: number;
+  /** Number of output tokens */
+  outputTokens?: number;
+  /** Number of cached input tokens */
+  cachedInputTokens?: number;
+  /** Number of cache creation input tokens */
+  cacheCreationInputTokens?: number;
+  /** Number of reasoning tokens */
+  reasoningTokens?: number;
+  /** Number of requests */
+  requestCount?: number;
+}
+
+export interface GatewaySpendReportResponse {
+  results: GatewaySpendReportRow[];
+}
+
+export class GatewaySpendReport {
+  constructor(private readonly config: GatewayConfig) {}
+
+  async getSpendReport(
+    params: GatewaySpendReportParams,
+  ): Promise<GatewaySpendReportResponse> {
+    try {
+      const baseUrl = new URL(this.config.baseURL);
+
+      const searchParams = new URLSearchParams();
+      searchParams.set('start_date', params.startDate);
+      searchParams.set('end_date', params.endDate);
+
+      if (params.groupBy) {
+        searchParams.set('group_by', params.groupBy);
+      }
+      if (params.datePart) {
+        searchParams.set('date_part', params.datePart);
+      }
+      if (params.userId) {
+        searchParams.set('user_id', params.userId);
+      }
+      if (params.model) {
+        searchParams.set('model', params.model);
+      }
+      if (params.provider) {
+        searchParams.set('provider', params.provider);
+      }
+      if (params.credentialType) {
+        searchParams.set('credential_type', params.credentialType);
+      }
+      if (params.tags && params.tags.length > 0) {
+        searchParams.set('tags', params.tags.join(','));
+      }
+
+      const { value } = await getFromApi({
+        url: `${baseUrl.origin}/v1/report?${searchParams.toString()}`,
+        headers: this.config.headers
+          ? await resolve(this.config.headers)
+          : undefined,
+        successfulResponseHandler: createJsonResponseHandler(
+          gatewaySpendReportResponseSchema,
+        ),
+        failedResponseHandler: createJsonErrorResponseHandler({
+          errorSchema: z.any(),
+          errorToMessage: data => data,
+        }),
+        fetch: this.config.fetch,
+      });
+
+      return value;
+    } catch (error) {
+      throw await asGatewayError(error);
+    }
+  }
+}
+
+const gatewaySpendReportResponseSchema = lazySchema(() =>
+  zodSchema(
+    z.object({
+      results: z.array(
+        z
+          .object({
+            day: z.string().optional(),
+            hour: z.string().optional(),
+            user: z.string().optional(),
+            model: z.string().optional(),
+            tag: z.string().optional(),
+            provider: z.string().optional(),
+            credential_type: z.enum(['byok', 'system']).optional(),
+            total_cost: z.number(),
+            market_cost: z.number().optional(),
+            input_tokens: z.number().optional(),
+            output_tokens: z.number().optional(),
+            cached_input_tokens: z.number().optional(),
+            cache_creation_input_tokens: z.number().optional(),
+            reasoning_tokens: z.number().optional(),
+            request_count: z.number().optional(),
+          })
+          .transform(
+            ({
+              credential_type,
+              total_cost,
+              market_cost,
+              input_tokens,
+              output_tokens,
+              cached_input_tokens,
+              cache_creation_input_tokens,
+              reasoning_tokens,
+              request_count,
+              ...rest
+            }) => ({
+              ...rest,
+              ...(credential_type !== undefined
+                ? { credentialType: credential_type }
+                : {}),
+              totalCost: total_cost,
+              ...(market_cost !== undefined ? { marketCost: market_cost } : {}),
+              ...(input_tokens !== undefined
+                ? { inputTokens: input_tokens }
+                : {}),
+              ...(output_tokens !== undefined
+                ? { outputTokens: output_tokens }
+                : {}),
+              ...(cached_input_tokens !== undefined
+                ? { cachedInputTokens: cached_input_tokens }
+                : {}),
+              ...(cache_creation_input_tokens !== undefined
+                ? { cacheCreationInputTokens: cache_creation_input_tokens }
+                : {}),
+              ...(reasoning_tokens !== undefined
+                ? { reasoningTokens: reasoning_tokens }
+                : {}),
+              ...(request_count !== undefined
+                ? { requestCount: request_count }
+                : {}),
+            }),
+          ),
+      ),
+    }),
+  ),
+);

package/src/gateway-video-model.ts

@@ -1,10 +1,10 @@
 import type {
-  Experimental_VideoModelV3,
-  Experimental_VideoModelV3CallOptions,
-  Experimental_VideoModelV3File,
-  Experimental_VideoModelV3VideoData,
-  SharedV3ProviderMetadata,
-  SharedV3Warning,
+  Experimental_VideoModelV4,
+  Experimental_VideoModelV4CallOptions,
+  Experimental_VideoModelV4File,
+  Experimental_VideoModelV4VideoData,
+  SharedV4ProviderMetadata,
+  SharedV4Warning,
 } from '@ai-sdk/provider';
 import { APICallError } from '@ai-sdk/provider';
 import {
@@ -21,8 +21,8 @@ import type { GatewayConfig } from './gateway-config';
 import { asGatewayError } from './errors';
 import { parseAuthMethod } from './errors/parse-auth-method';
 
-export class GatewayVideoModel implements Experimental_VideoModelV3 {
-  readonly specificationVersion = 'v3' as const;
+export class GatewayVideoModel implements Experimental_VideoModelV4 {
+  readonly specificationVersion = 'v4' as const;
   // Set a very large number to prevent client-side splitting of requests
   readonly maxVideosPerCall = Number.MAX_SAFE_INTEGER;
 
@@ -50,17 +50,19 @@ export class GatewayVideoModel implements Experimental_VideoModelV3 {
     providerOptions,
     headers,
     abortSignal,
-  }: Experimental_VideoModelV3CallOptions): Promise<{
-    videos: Array<Experimental_VideoModelV3VideoData>;
-    warnings: Array<SharedV3Warning>;
-    providerMetadata?: SharedV3ProviderMetadata;
+  }: Experimental_VideoModelV4CallOptions): Promise<{
+    videos: Array<Experimental_VideoModelV4VideoData>;
+    warnings: Array<SharedV4Warning>;
+    providerMetadata?: SharedV4ProviderMetadata;
     response: {
       timestamp: Date;
       modelId: string;
       headers: Record<string, string> | undefined;
     };
   }> {
-    const resolvedHeaders = await resolve(this.config.headers());
+    const resolvedHeaders = this.config.headers
+      ? await resolve(this.config.headers)
+      : undefined;
     try {
       const { responseHeaders, value: responseBody } = await postJsonToApi({
         url: this.getUrl(),
@@ -170,7 +172,7 @@ export class GatewayVideoModel implements Experimental_VideoModelV3 {
         videos: responseBody.videos,
         warnings: responseBody.warnings ?? [],
         providerMetadata:
-          responseBody.providerMetadata as SharedV3ProviderMetadata,
+          responseBody.providerMetadata as SharedV4ProviderMetadata,
         response: {
           timestamp: new Date(),
           modelId: this.modelId,
@@ -178,7 +180,10 @@ export class GatewayVideoModel implements Experimental_VideoModelV3 {
         },
       };
     } catch (error) {
-      throw await asGatewayError(error, await parseAuthMethod(resolvedHeaders));
+      throw await asGatewayError(
+        error,
+        await parseAuthMethod(resolvedHeaders ?? {}),
+      );
     }
   }
 
@@ -188,13 +193,13 @@ export class GatewayVideoModel implements Experimental_VideoModelV3 {
 
   private getModelConfigHeaders() {
     return {
-      'ai-video-model-specification-version': '3',
+      'ai-video-model-specification-version': '4',
       'ai-model-id': this.modelId,
     };
   }
 }
 
-function maybeEncodeVideoFile(file: Experimental_VideoModelV3File) {
+function maybeEncodeVideoFile(file: Experimental_VideoModelV4File) {
   if (file.type === 'file' && file.data instanceof Uint8Array) {
     return {
       ...file,