@ai-sdk/gateway 4.0.0-beta.4 → 4.0.0-beta.41

package/src/gateway-generation-info.ts

@@ -0,0 +1,147 @@
+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 GatewayGenerationInfoParams {
+  /** The generation ID to look up (format: gen_<ulid>) */
+  id: string;
+}
+
+export interface GatewayGenerationInfo {
+  /** The generation ID */
+  id: string;
+  /** Total cost in USD */
+  totalCost: number;
+  /** Upstream inference cost in USD (BYOK only) */
+  upstreamInferenceCost: number;
+  /** Usage cost in USD (same as totalCost) */
+  usage: number;
+  /** ISO 8601 timestamp when the generation was created */
+  createdAt: string;
+  /** Model identifier */
+  model: string;
+  /** Whether BYOK credentials were used */
+  isByok: boolean;
+  /** Provider that served this generation */
+  providerName: string;
+  /** Whether streaming was used */
+  streamed: boolean;
+  /** Finish reason (e.g. 'stop') */
+  finishReason: string;
+  /** Time to first token in milliseconds */
+  latency: number;
+  /** Total generation time in milliseconds */
+  generationTime: number;
+  /** Number of prompt tokens */
+  promptTokens: number;
+  /** Number of completion tokens */
+  completionTokens: number;
+  /** Reasoning tokens used */
+  reasoningTokens: number;
+  /** Cached tokens used */
+  cachedTokens: number;
+  /** Cache creation input tokens */
+  cacheCreationTokens: number;
+  /** Billable web search calls */
+  billableWebSearchCalls: number;
+}
+
+export class GatewayGenerationInfoFetcher {
+  constructor(private readonly config: GatewayConfig) {}
+
+  async getGenerationInfo(
+    params: GatewayGenerationInfoParams,
+  ): Promise<GatewayGenerationInfo> {
+    try {
+      const baseUrl = new URL(this.config.baseURL);
+
+      const { value } = await getFromApi({
+        url: `${baseUrl.origin}/v1/generation?id=${encodeURIComponent(params.id)}`,
+        headers: await resolve(this.config.headers()),
+        successfulResponseHandler: createJsonResponseHandler(
+          gatewayGenerationInfoResponseSchema,
+        ),
+        failedResponseHandler: createJsonErrorResponseHandler({
+          errorSchema: z.any(),
+          errorToMessage: data => data,
+        }),
+        fetch: this.config.fetch,
+      });
+
+      return value;
+    } catch (error) {
+      throw await asGatewayError(error);
+    }
+  }
+}
+
+const gatewayGenerationInfoResponseSchema = lazySchema(() =>
+  zodSchema(
+    z
+      .object({
+        data: z
+          .object({
+            id: z.string(),
+            total_cost: z.number(),
+            upstream_inference_cost: z.number(),
+            usage: z.number(),
+            created_at: z.string(),
+            model: z.string(),
+            is_byok: z.boolean(),
+            provider_name: z.string(),
+            streamed: z.boolean(),
+            finish_reason: z.string(),
+            latency: z.number(),
+            generation_time: z.number(),
+            native_tokens_prompt: z.number(),
+            native_tokens_completion: z.number(),
+            native_tokens_reasoning: z.number(),
+            native_tokens_cached: z.number(),
+            native_tokens_cache_creation: z.number(),
+            billable_web_search_calls: z.number(),
+          })
+          .transform(
+            ({
+              total_cost,
+              upstream_inference_cost,
+              created_at,
+              is_byok,
+              provider_name,
+              finish_reason,
+              generation_time,
+              native_tokens_prompt,
+              native_tokens_completion,
+              native_tokens_reasoning,
+              native_tokens_cached,
+              native_tokens_cache_creation,
+              billable_web_search_calls,
+              ...rest
+            }) => ({
+              ...rest,
+              totalCost: total_cost,
+              upstreamInferenceCost: upstream_inference_cost,
+              createdAt: created_at,
+              isByok: is_byok,
+              providerName: provider_name,
+              finishReason: finish_reason,
+              generationTime: generation_time,
+              promptTokens: native_tokens_prompt,
+              completionTokens: native_tokens_completion,
+              reasoningTokens: native_tokens_reasoning,
+              cachedTokens: native_tokens_cached,
+              cacheCreationTokens: native_tokens_cache_creation,
+              billableWebSearchCalls: billable_web_search_calls,
+            }),
+          ),
+      })
+      .transform(({ data }) => data),
+  ),
+);

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

@@ -1,4 +1,9 @@
 export type GatewayImageModelId =
+  | 'bfl/flux-2-flex'
+  | 'bfl/flux-2-klein-4b'
+  | 'bfl/flux-2-klein-9b'
+  | 'bfl/flux-2-max'
+  | 'bfl/flux-2-pro'
   | 'bfl/flux-kontext-max'
   | 'bfl/flux-kontext-pro'
   | 'bfl/flux-pro-1.0-fill'
@@ -10,6 +15,7 @@ export type GatewayImageModelId =
   | 'openai/gpt-image-1'
   | 'openai/gpt-image-1-mini'
   | 'openai/gpt-image-1.5'
+  | 'prodia/flux-fast-schnell'
   | 'recraft/recraft-v2'
   | 'recraft/recraft-v3'
   | 'recraft/recraft-v4'

package/src/gateway-image-model.ts

@@ -1,7 +1,7 @@
 import type {
-  ImageModelV3,
-  ImageModelV3File,
-  ImageModelV3ProviderMetadata,
+  ImageModelV4,
+  ImageModelV4File,
+  ImageModelV4ProviderMetadata,
 } from '@ai-sdk/provider';
 import {
   combineHeaders,
@@ -17,8 +17,8 @@ import type { GatewayConfig } from './gateway-config';
 import { asGatewayError } from './errors';
 import { parseAuthMethod } from './errors/parse-auth-method';
 
-export class GatewayImageModel implements ImageModelV3 {
-  readonly specificationVersion = 'v3' as const;
+export class GatewayImageModel implements ImageModelV4 {
+  readonly specificationVersion = 'v4' as const;
   // Set a very large number to prevent client-side splitting of requests
   readonly maxImagesPerCall = Number.MAX_SAFE_INTEGER;
 
@@ -45,8 +45,8 @@ export class GatewayImageModel implements ImageModelV3 {
     providerOptions,
     headers,
     abortSignal,
-  }: Parameters<ImageModelV3['doGenerate']>[0]): Promise<
-    Awaited<ReturnType<ImageModelV3['doGenerate']>>
+  }: Parameters<ImageModelV4['doGenerate']>[0]): Promise<
+    Awaited<ReturnType<ImageModelV4['doGenerate']>>
   > {
     const resolvedHeaders = await resolve(this.config.headers());
     try {
@@ -89,7 +89,7 @@ export class GatewayImageModel implements ImageModelV3 {
         images: responseBody.images, // Always base64 strings from server
         warnings: responseBody.warnings ?? [],
         providerMetadata:
-          responseBody.providerMetadata as ImageModelV3ProviderMetadata,
+          responseBody.providerMetadata as ImageModelV4ProviderMetadata,
         response: {
           timestamp: new Date(),
           modelId: this.modelId,
@@ -114,13 +114,13 @@ export class GatewayImageModel implements ImageModelV3 {
 
   private getModelConfigHeaders() {
     return {
-      'ai-image-model-specification-version': '3',
+      'ai-image-model-specification-version': '4',
       'ai-model-id': this.modelId,
     };
   }
 }
 
-function maybeEncodeImageFile(file: ImageModelV3File) {
+function maybeEncodeImageFile(file: ImageModelV4File) {
   if (file.type === 'file' && file.data instanceof Uint8Array) {
     return {
       ...file,

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

@@ -17,6 +17,7 @@ export type GatewayModelId =
   | 'alibaba/qwen3-vl-thinking'
   | 'alibaba/qwen3.5-flash'
   | 'alibaba/qwen3.5-plus'
+  | 'alibaba/qwen3.6-plus'
   | 'amazon/nova-2-lite'
   | 'amazon/nova-lite'
   | 'amazon/nova-micro'
@@ -24,8 +25,6 @@ export type GatewayModelId =
   | 'anthropic/claude-3-haiku'
   | 'anthropic/claude-3-opus'
   | 'anthropic/claude-3.5-haiku'
-  | 'anthropic/claude-3.5-sonnet'
-  | 'anthropic/claude-3.5-sonnet-20240620'
   | 'anthropic/claude-3.7-sonnet'
   | 'anthropic/claude-haiku-4.5'
   | 'anthropic/claude-opus-4'
@@ -36,6 +35,7 @@ export type GatewayModelId =
   | 'anthropic/claude-sonnet-4.5'
   | 'anthropic/claude-sonnet-4.6'
   | 'arcee-ai/trinity-large-preview'
+  | 'arcee-ai/trinity-large-thinking'
   | 'arcee-ai/trinity-mini'
   | 'bytedance/seed-1.6'
   | 'bytedance/seed-1.8'
@@ -51,8 +51,6 @@ export type GatewayModelId =
   | 'google/gemini-2.5-flash'
   | 'google/gemini-2.5-flash-image'
   | 'google/gemini-2.5-flash-lite'
-  | 'google/gemini-2.5-flash-lite-preview-09-2025'
-  | 'google/gemini-2.5-flash-preview-09-2025'
   | 'google/gemini-2.5-pro'
   | 'google/gemini-3-flash'
   | 'google/gemini-3-pro-image'
@@ -60,11 +58,14 @@ export type GatewayModelId =
   | 'google/gemini-3.1-flash-image-preview'
   | 'google/gemini-3.1-flash-lite-preview'
   | 'google/gemini-3.1-pro-preview'
+  | 'google/gemma-4-26b-a4b-it'
+  | 'google/gemma-4-31b-it'
   | 'inception/mercury-2'
   | 'inception/mercury-coder-small'
   | 'kwaipilot/kat-coder-pro-v1'
+  | 'kwaipilot/kat-coder-pro-v2'
   | 'meituan/longcat-flash-chat'
-  | 'meituan/longcat-flash-thinking'
+  | 'meituan/longcat-flash-thinking-2601'
   | 'meta/llama-3.1-70b'
   | 'meta/llama-3.1-8b'
   | 'meta/llama-3.2-11b'
@@ -79,6 +80,8 @@ export type GatewayModelId =
   | 'minimax/minimax-m2.1-lightning'
   | 'minimax/minimax-m2.5'
   | 'minimax/minimax-m2.5-highspeed'
+  | 'minimax/minimax-m2.7'
+  | 'minimax/minimax-m2.7-highspeed'
   | 'mistral/codestral'
   | 'mistral/devstral-2'
   | 'mistral/devstral-small'
@@ -104,9 +107,9 @@ export type GatewayModelId =
   | 'morph/morph-v3-fast'
   | 'morph/morph-v3-large'
   | 'nvidia/nemotron-3-nano-30b-a3b'
+  | 'nvidia/nemotron-3-super-120b-a12b'
   | 'nvidia/nemotron-nano-12b-v2-vl'
   | 'nvidia/nemotron-nano-9b-v2'
-  | 'openai/codex-mini'
   | 'openai/gpt-3.5-turbo'
   | 'openai/gpt-3.5-turbo-instruct'
   | 'openai/gpt-4-turbo'
@@ -134,6 +137,8 @@ export type GatewayModelId =
   | 'openai/gpt-5.3-chat'
   | 'openai/gpt-5.3-codex'
   | 'openai/gpt-5.4'
+  | 'openai/gpt-5.4-mini'
+  | 'openai/gpt-5.4-nano'
   | 'openai/gpt-5.4-pro'
   | 'openai/gpt-oss-120b'
   | 'openai/gpt-oss-20b'
@@ -146,12 +151,8 @@ export type GatewayModelId =
   | 'openai/o4-mini'
   | 'perplexity/sonar'
   | 'perplexity/sonar-pro'
-  | 'perplexity/sonar-reasoning'
   | 'perplexity/sonar-reasoning-pro'
   | 'prime-intellect/intellect-3'
-  | 'vercel/v0-1.0-md'
-  | 'vercel/v0-1.5-md'
-  | 'xai/grok-2-vision'
   | 'xai/grok-3'
   | 'xai/grok-3-fast'
   | 'xai/grok-3-mini'
@@ -161,8 +162,15 @@ export type GatewayModelId =
   | 'xai/grok-4-fast-reasoning'
   | 'xai/grok-4.1-fast-non-reasoning'
   | 'xai/grok-4.1-fast-reasoning'
+  | 'xai/grok-4.20-multi-agent'
+  | 'xai/grok-4.20-multi-agent-beta'
+  | 'xai/grok-4.20-non-reasoning'
+  | 'xai/grok-4.20-non-reasoning-beta'
+  | 'xai/grok-4.20-reasoning'
+  | 'xai/grok-4.20-reasoning-beta'
   | 'xai/grok-code-fast-1'
   | 'xiaomi/mimo-v2-flash'
+  | 'xiaomi/mimo-v2-pro'
   | 'zai/glm-4.5'
   | 'zai/glm-4.5-air'
   | 'zai/glm-4.5v'
@@ -173,4 +181,7 @@ export type GatewayModelId =
   | 'zai/glm-4.7-flash'
   | 'zai/glm-4.7-flashx'
   | 'zai/glm-5'
+  | 'zai/glm-5-turbo'
+  | 'zai/glm-5.1'
+  | 'zai/glm-5v-turbo'
   | (string & {});

package/src/gateway-language-model.ts

@@ -1,11 +1,11 @@
 import type {
-  LanguageModelV3,
-  LanguageModelV3CallOptions,
-  SharedV3Warning,
-  LanguageModelV3FilePart,
-  LanguageModelV3StreamPart,
-  LanguageModelV3GenerateResult,
-  LanguageModelV3StreamResult,
+  LanguageModelV4,
+  LanguageModelV4CallOptions,
+  SharedV4Warning,
+  LanguageModelV4FilePart,
+  LanguageModelV4StreamPart,
+  LanguageModelV4GenerateResult,
+  LanguageModelV4StreamResult,
 } from '@ai-sdk/provider';
 import {
   combineHeaders,
@@ -28,8 +28,8 @@ type GatewayChatConfig = GatewayConfig & {
   o11yHeaders: Resolvable<Record<string, string>>;
 };
 
-export class GatewayLanguageModel implements LanguageModelV3 {
-  readonly specificationVersion = 'v3';
+export class GatewayLanguageModel implements LanguageModelV4 {
+  readonly specificationVersion = 'v4';
   readonly supportedUrls = { '*/*': [/.*/] };
 
   constructor(
@@ -41,7 +41,7 @@ export class GatewayLanguageModel implements LanguageModelV3 {
     return this.config.provider;
   }
 
-  private async getArgs(options: LanguageModelV3CallOptions) {
+  private async getArgs(options: LanguageModelV4CallOptions) {
     const { abortSignal: _abortSignal, ...optionsWithoutSignal } = options;
 
     return {
@@ -51,8 +51,8 @@ export class GatewayLanguageModel implements LanguageModelV3 {
   }
 
   async doGenerate(
-    options: LanguageModelV3CallOptions,
-  ): Promise<LanguageModelV3GenerateResult> {
+    options: LanguageModelV4CallOptions,
+  ): Promise<LanguageModelV4GenerateResult> {
     const { args, warnings } = await this.getArgs(options);
     const { abortSignal } = options;
 
@@ -93,8 +93,8 @@ export class GatewayLanguageModel implements LanguageModelV3 {
   }
 
   async doStream(
-    options: LanguageModelV3CallOptions,
-  ): Promise<LanguageModelV3StreamResult> {
+    options: LanguageModelV4CallOptions,
+  ): Promise<LanguageModelV4StreamResult> {
     const { args, warnings } = await this.getArgs(options);
     const { abortSignal } = options;
 
@@ -122,8 +122,8 @@ export class GatewayLanguageModel implements LanguageModelV3 {
       return {
         stream: response.pipeThrough(
           new TransformStream<
-            ParseResult<LanguageModelV3StreamPart>,
-            LanguageModelV3StreamPart
+            ParseResult<LanguageModelV4StreamPart>,
+            LanguageModelV4StreamPart
           >({
             start(controller) {
               if (warnings.length > 0) {
@@ -177,11 +177,11 @@ export class GatewayLanguageModel implements LanguageModelV3 {
    * @param options - The options to encode.
    * @returns The options with the file parts encoded.
    */
-  private maybeEncodeFileParts(options: LanguageModelV3CallOptions) {
+  private maybeEncodeFileParts(options: LanguageModelV4CallOptions) {
     for (const message of options.prompt) {
       for (const part of message.content) {
         if (this.isFilePart(part)) {
-          const filePart = part as LanguageModelV3FilePart;
+          const filePart = part as LanguageModelV4FilePart;
           // If the file part is a URL it will get cleanly converted to a string.
           // If it's a binary file attachment we convert it to a data url.
           // In either case, server-side we should only ever see URLs as strings.
@@ -204,7 +204,7 @@ export class GatewayLanguageModel implements LanguageModelV3 {
 
   private getModelConfigHeaders(modelId: string, streaming: boolean) {
     return {
-      'ai-language-model-specification-version': '3',
+      'ai-language-model-specification-version': '4',
       'ai-language-model-id': modelId,
       'ai-language-model-streaming': String(streaming),
     };

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({
       /**
@@ -53,12 +53,33 @@ 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 +96,4 @@ const gatewayLanguageModelOptions = lazySchema(() =>
   ),
 );
 
-export type GatewayLanguageModelOptions = InferSchema<
-  typeof gatewayLanguageModelOptions
->;
+export type GatewayProviderOptions = InferSchema<typeof gatewayProviderOptions>;