feat(workers-ai-provider): support AI SDK v7

.changeset/witty-seas-attend.md

@@ -0,0 +1,7 @@
+---
+"workers-ai-provider": major
+---
+
+Support the AI SDK v7. Peer dependencies now require `ai@^7`, `@ai-sdk/provider@^4`, and (for the AI Gateway sub-path plugins) `@ai-sdk/openai@^4`, `@ai-sdk/anthropic@^4`, and `@ai-sdk/google@^4`. AI SDK v6 is no longer supported.
+
+The Workers AI models (chat, embeddings, image, transcription, speech, reranking) continue to implement the `*ModelV3` specs, which AI SDK v7 still accepts unchanged, so their behavior is identical. The AI Gateway delegate — which wraps the third-party `@ai-sdk/*` providers routed through Gateway — is migrated from the `LanguageModelV3` spec to `LanguageModelV4` to match those providers on v7. No runtime behavior changes; the delegate remains a pass-through over the underlying provider models.

packages/workers-ai-provider/package.json

@@ -65,21 +65,21 @@
 		"type-check": "tsc --noEmit"
 	},
 	"devDependencies": {
-		"@ai-sdk/anthropic": "^3.0.84",
-		"@ai-sdk/google": "^3.0.0",
-		"@ai-sdk/openai": "^3.0.71",
-		"@ai-sdk/provider": "^3.0.10",
+		"@ai-sdk/anthropic": "^4.0.0",
+		"@ai-sdk/google": "^4.0.0",
+		"@ai-sdk/openai": "^4.0.0",
+		"@ai-sdk/provider": "^4.0.0",
 		"@cloudflare/gateway-core": "workspace:*",
 		"@cloudflare/workers-types": "^4.20260628.1",
-		"ai": "^6.0.204",
+		"ai": "^7.0.0",
 		"zod": "^4.4.3"
 	},
 	"peerDependencies": {
-		"@ai-sdk/anthropic": "^3.0.0",
-		"@ai-sdk/google": "^3.0.0",
-		"@ai-sdk/openai": "^3.0.0",
-		"@ai-sdk/provider": "^3.0.0",
-		"ai": "^6.0.0"
+		"@ai-sdk/anthropic": "^4.0.0",
+		"@ai-sdk/google": "^4.0.0",
+		"@ai-sdk/openai": "^4.0.0",
+		"@ai-sdk/provider": "^4.0.0",
+		"ai": "^7.0.0"
 	},
 	"peerDependenciesMeta": {
 		"@ai-sdk/anthropic": {

packages/workers-ai-provider/src/client-fallback.ts

@@ -1,8 +1,8 @@
 import type {
-	LanguageModelV3,
-	LanguageModelV3CallOptions,
-	LanguageModelV3GenerateResult,
-	LanguageModelV3StreamResult,
+	LanguageModelV4,
+	LanguageModelV4CallOptions,
+	LanguageModelV4GenerateResult,
+	LanguageModelV4StreamResult,
 } from "@ai-sdk/provider";
 import { type FallbackAttempt, WorkersAIFallbackError, WorkersAIGatewayError } from "./errors";
 import type { Transport } from "./gateway-delegate";
@@ -12,7 +12,7 @@ export interface FallbackLeg {
 	/** The model slug this leg dispatches. */
 	slug: string;
 	/** The built AI SDK model. */
-	model: LanguageModelV3;
+	model: LanguageModelV4;
 	/** Transport the leg uses. */
 	transport: Transport;
 }
@@ -26,13 +26,13 @@ export interface FallbackLeg {
  * produced a stream). Errors that surface *mid-stream* — after content has
  * already been emitted — are not recoverable here and propagate as-is.
  */
-export function createClientFallbackModel(legs: FallbackLeg[]): LanguageModelV3 {
+export function createClientFallbackModel(legs: FallbackLeg[]): LanguageModelV4 {
 	if (legs.length === 0) {
 		throw new Error("createClientFallbackModel requires at least one model leg.");
 	}
 	const primary = legs[0].model;
 
-	async function attempt<T>(run: (model: LanguageModelV3) => PromiseLike<T>): Promise<T> {
+	async function attempt<T>(run: (model: LanguageModelV4) => PromiseLike<T>): Promise<T> {
 		const attempts: FallbackAttempt[] = [];
 		for (const leg of legs) {
 			try {
@@ -54,16 +54,16 @@ export function createClientFallbackModel(legs: FallbackLeg[]): LanguageModelV3
 	}
 
 	return {
-		specificationVersion: "v3",
+		specificationVersion: "v4",
 		provider: primary.provider,
 		modelId: primary.modelId,
 		supportedUrls: primary.supportedUrls,
 		doGenerate(
-			options: LanguageModelV3CallOptions,
-		): PromiseLike<LanguageModelV3GenerateResult> {
+			options: LanguageModelV4CallOptions,
+		): PromiseLike<LanguageModelV4GenerateResult> {
 			return attempt((m) => m.doGenerate(options));
 		},
-		doStream(options: LanguageModelV3CallOptions): PromiseLike<LanguageModelV3StreamResult> {
+		doStream(options: LanguageModelV4CallOptions): PromiseLike<LanguageModelV4StreamResult> {
 			return attempt((m) => m.doStream(options));
 		},
 	};

packages/workers-ai-provider/src/gateway-delegate.ts

@@ -1,4 +1,4 @@
-import type { LanguageModelV3, LanguageModelV3CallOptions } from "@ai-sdk/provider";
+import type { LanguageModelV4, LanguageModelV4CallOptions } from "@ai-sdk/provider";
 import {
 	asText,
 	buildGatewayEntry,
@@ -154,7 +154,7 @@ export interface ProviderPlugin {
 		modelId: string;
 		fetch: typeof globalThis.fetch;
 		baseURL?: string;
-	}): LanguageModelV3;
+	}): LanguageModelV4;
 }
 
 // ---------------------------------------------------------------------------
@@ -401,7 +401,7 @@ export interface GatewayDelegateConfig {
 }
 
 export interface GatewayDelegate {
-	(slug: string, options?: DelegateCallOptions): LanguageModelV3;
+	(slug: string, options?: DelegateCallOptions): LanguageModelV4;
 }
 
 /**
@@ -431,7 +431,7 @@ export function createGatewayDelegate(config: GatewayDelegateConfig): GatewayDel
 	const buildOne = (
 		slug: string,
 		options: DelegateCallOptions,
-	): { model: LanguageModelV3; transport: Transport } => {
+	): { model: LanguageModelV4; transport: Transport } => {
 		const parsed = parseSlug(slug);
 		const info = resolveProvider(slug, parsed);
 
@@ -816,7 +816,7 @@ function makeServerFallbackModel(params: {
 	opts: DelegateCallOptions;
 	selection: Selection;
 	callOptions: DelegateCallOptions;
-}): LanguageModelV3 {
+}): LanguageModelV4 {
 	const { binding, gatewayId, gatewayOptions, legs, opts, selection, callOptions } = params;
 	const first = legs[0]!;
 
@@ -834,7 +834,7 @@ function makeServerFallbackModel(params: {
 
 	const dispatch = async (
 		method: "doGenerate" | "doStream",
-		options: LanguageModelV3CallOptions,
+		options: LanguageModelV4CallOptions,
 	): Promise<unknown> => {
 		// 1) Capture each leg's native request without hitting the network.
 		const entries: GatewayEntry[] = [];
@@ -854,7 +854,7 @@ function makeServerFallbackModel(params: {
 				...(leg.info.baseURL ? { baseURL: leg.info.baseURL } : {}),
 			});
 			try {
-				await (model[method] as (o: LanguageModelV3CallOptions) => Promise<unknown>)(
+				await (model[method] as (o: LanguageModelV4CallOptions) => Promise<unknown>)(
 					options,
 				);
 			} catch (e) {
@@ -905,21 +905,21 @@ function makeServerFallbackModel(params: {
 			fetch: (async () => resp) as typeof globalThis.fetch,
 			...(winner.info.baseURL ? { baseURL: winner.info.baseURL } : {}),
 		});
-		return (winnerModel[method] as (o: LanguageModelV3CallOptions) => Promise<unknown>)(
+		return (winnerModel[method] as (o: LanguageModelV4CallOptions) => Promise<unknown>)(
 			options,
 		);
 	};
 
 	return {
-		specificationVersion: "v3",
+		specificationVersion: "v4",
 		provider: refModel.provider,
 		modelId: refModel.modelId,
 		supportedUrls: refModel.supportedUrls,
 		doGenerate(options) {
-			return dispatch("doGenerate", options) as ReturnType<LanguageModelV3["doGenerate"]>;
+			return dispatch("doGenerate", options) as ReturnType<LanguageModelV4["doGenerate"]>;
 		},
 		doStream(options) {
-			return dispatch("doStream", options) as ReturnType<LanguageModelV3["doStream"]>;
+			return dispatch("doStream", options) as ReturnType<LanguageModelV4["doStream"]>;
 		},
-	} as LanguageModelV3;
+	} as LanguageModelV4;
 }

packages/workers-ai-provider/src/index.ts

@@ -508,7 +508,7 @@ export function createWorkersAI(options: WorkersAISettings): WorkersAI {
 					"catalog",
 				);
 			}
-			// The delegate returns a `LanguageModelV3` built by the configured plugin.
+			// The delegate returns a `LanguageModelV4` built by the configured plugin.
 			// It's structurally compatible with the AI SDK consumers this provider is
 			// used with; the cast keeps the public return type unchanged.
 			return getDelegate(modelId)(