fix(ai): stop retrying Z.ai Responses model and permission rejections

packages/ai/src/provider-error.ts

@@ -114,7 +114,8 @@ const QUOTA_CODES = new Set([
   "creditlimitexceeded",
 ])
 // Google reports an invalid API key as HTTP 400 INVALID_ARGUMENT with this `details[].reason`.
-const AUTH_CODES = new Set(["authentication_error", "permission_error", "api_key_invalid"])
+// Z.ai's Responses API reports account and plan rejections mid-stream as `permission_denied`.
+const AUTH_CODES = new Set(["authentication_error", "permission_error", "permission_denied", "api_key_invalid"])
 const SERVER_CODES = new Set([
   "api_error",
   "internal_error",
@@ -128,6 +129,7 @@ const SERVER_CODES = new Set([
 ])
 // `invalid_request` is the Vercel AI Gateway's code for an upstream request rejection.
 const INVALID_REQUEST_CODES = new Set([
+  "model_not_found",
   "invalid_prompt",
   "invalid_request",
   "invalid_request_error",

packages/ai/test/provider-error.test.ts

@@ -422,6 +422,19 @@ describe("provider error rawBody classification", () => {
     expect(
       classifyProviderFailure({ message: "Request failed", rawBody: '{"error":{"code":"insufficient_quota"}}' })._tag,
     ).toBe("QuotaExceeded")
+    // Z.ai Responses stream rejections
+    expect(
+      classifyProviderFailure({
+        message: "Unknown Model, please check the model code.",
+        rawBody: '{"type":"response.failed","response":{"error":{"code":"model_not_found"}}}',
+      })._tag,
+    ).toBe("InvalidRequest")
+    expect(
+      classifyProviderFailure({
+        message: "Your GLM Coding Plan package has expired and is temporarily unavailable.",
+        rawBody: '{"type":"response.failed","response":{"error":{"code":"permission_denied"}}}',
+      })._tag,
+    ).toBe("Authentication")
   })
 })