From e97a7027932ae97ebc206032b4b0e3397bd026df Mon Sep 17 00:00:00 2001
From: mikemikimike <13286568797@163.com>
Date: Sun, 23 Aug 2026 00:29:53 +0800
Subject: [PATCH 1/2] fix(workers-ai-provider): route gpt-5.6 through responses

---
 .changeset/fix-gpt-56-responses.md            |  5 ++++
 packages/workers-ai-provider/src/openai.ts    | 10 ++++---
 .../workers-ai-provider/test/openai.test.ts   | 26 +++++++++++++++++++
 3 files changed, 37 insertions(+), 4 deletions(-)
 create mode 100644 .changeset/fix-gpt-56-responses.md
 create mode 100644 packages/workers-ai-provider/test/openai.test.ts

diff --git a/.changeset/fix-gpt-56-responses.md b/.changeset/fix-gpt-56-responses.md
new file mode 100644
index 0000000000..bc55888035
--- /dev/null
+++ b/.changeset/fix-gpt-56-responses.md
@@ -0,0 +1,5 @@
+---
+"workers-ai-provider": patch
+---
+
+Route OpenAI gpt-5.6 models through the Responses API.
diff --git a/packages/workers-ai-provider/src/openai.ts b/packages/workers-ai-provider/src/openai.ts
index 7312f64e68..14e326d99b 100644
--- a/packages/workers-ai-provider/src/openai.ts
+++ b/packages/workers-ai-provider/src/openai.ts
@@ -11,9 +11,9 @@ import type { ProviderPlugin } from "./gateway-delegate";
  *
  * Requires `@ai-sdk/openai` (an optional peer dependency — install it yourself).
  *
- * Uses `.chat()` (Chat Completions) deliberately: the bare `openai()` call
- * (AI SDK v6+) defaults to the Responses API, which the AI Gateway run catalog
- * does not serve.
+ * Uses `.chat()` (Chat Completions) for the catalog's chat-completions models.
+ * The gpt-5.6 class only accepts the Responses API, so those models use
+ * `.responses()` instead.
  */
 export const openai: ProviderPlugin = {
 	wireFormat: "openai",
@@ -22,5 +22,7 @@ export const openai: ProviderPlugin = {
 		// and the delegate strips the Authorization header on the gateway path.
 		// baseURL (set by the registry for non-OpenAI openai-wire providers) makes
 		// the generated URL host-strip to the right gateway-native endpoint.
-		createOpenAI({ apiKey: "unused", fetch, ...(baseURL ? { baseURL } : {}) }).chat(modelId),
+		createOpenAI({ apiKey: "unused", fetch, ...(baseURL ? { baseURL } : {}) })[
+			/^gpt-5\.6(?:-|$)/.test(modelId) ? "responses" : "chat"
+		](modelId),
 };
diff --git a/packages/workers-ai-provider/test/openai.test.ts b/packages/workers-ai-provider/test/openai.test.ts
new file mode 100644
index 0000000000..35d180443e
--- /dev/null
+++ b/packages/workers-ai-provider/test/openai.test.ts
@@ -0,0 +1,26 @@
+import { describe, expect, it, vi } from "vitest";
+import { openai } from "../src/openai";
+
+async function requestedUrl(modelId: string) {
+	const fetch = vi.fn(async (input: RequestInfo | URL) => {
+		throw new Error(String(input));
+	});
+
+	const model = openai.create({
+		modelId,
+		fetch: fetch as typeof globalThis.fetch,
+	});
+
+	await expect(model.doGenerate({ prompt: [] } as never)).rejects.toThrow();
+	return String(fetch.mock.calls[0]?.[0]);
+}
+
+describe("openai provider plugin", () => {
+	it("uses the Responses API for gpt-5.6 models", async () => {
+		expect(await requestedUrl("gpt-5.6-luna")).toContain("/responses");
+	});
+
+	it("keeps Chat Completions for ordinary OpenAI models", async () => {
+		expect(await requestedUrl("gpt-5.4-mini")).toContain("/chat/completions");
+	});
+});

From 570dc20583cfc9ab979b409c0b31af98a9a7127e Mon Sep 17 00:00:00 2001
From: mikemikimike <13286568797@163.com>
Date: Sun, 23 Aug 2026 00:37:11 +0800
Subject: [PATCH 2/2] test(workers-ai-provider): cover gpt-5.6 routing
 boundaries

---
 .../workers-ai-provider/test/openai.test.ts     | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

diff --git a/packages/workers-ai-provider/test/openai.test.ts b/packages/workers-ai-provider/test/openai.test.ts
index 35d180443e..9ee9c48367 100644
--- a/packages/workers-ai-provider/test/openai.test.ts
+++ b/packages/workers-ai-provider/test/openai.test.ts
@@ -1,7 +1,7 @@
 import { describe, expect, it, vi } from "vitest";
 import { openai } from "../src/openai";
 
-async function requestedUrl(modelId: string) {
+async function requestedUrl(modelId: string, baseURL?: string) {
 	const fetch = vi.fn(async (input: RequestInfo | URL) => {
 		throw new Error(String(input));
 	});
@@ -9,6 +9,7 @@ async function requestedUrl(modelId: string) {
 	const model = openai.create({
 		modelId,
 		fetch: fetch as typeof globalThis.fetch,
+		...(baseURL ? { baseURL } : {}),
 	});
 
 	await expect(model.doGenerate({ prompt: [] } as never)).rejects.toThrow();
@@ -17,10 +18,20 @@ async function requestedUrl(modelId: string) {
 
 describe("openai provider plugin", () => {
 	it("uses the Responses API for gpt-5.6 models", async () => {
-		expect(await requestedUrl("gpt-5.6-luna")).toContain("/responses");
+		for (const modelId of ["gpt-5.6", "gpt-5.6-luna", "gpt-5.6-mini"]) {
+			expect(await requestedUrl(modelId)).toContain("/responses");
+		}
 	});
 
 	it("keeps Chat Completions for ordinary OpenAI models", async () => {
-		expect(await requestedUrl("gpt-5.4-mini")).toContain("/chat/completions");
+		for (const modelId of ["gpt-5.4-mini", "gpt-5.5"]) {
+			expect(await requestedUrl(modelId)).toContain("/chat/completions");
+		}
+	});
+
+	it("preserves the Responses endpoint when a Gateway base URL is configured", async () => {
+		expect(await requestedUrl("gpt-5.6-luna", "https://gateway.example/v1")).toBe(
+			"https://gateway.example/v1/responses",
+		);
 	});
 });