fix(codex): Test probe uses a ChatGPT-account-supported model (#7521) (#7524)

The connection Test button always reported success for a Codex connection
backed by a ChatGPT account: the probe sent `gpt-5.3-codex`, a codex-only
model the ChatGPT-account backend rejects outright with a 400 — the same
status the probe treats as 'auth accepted, body invalid'. A bad token and a
good token both came back 400, so Test could never fail on a bad token.

Probe with `gpt-5.5` (confirmed served for ChatGPT-account sessions via live
VPS test 2026-07-16) instead; `input: []` still yields the intended 400 for a
good token, 401/403 for a bad one.

Live verification (VPS): gpt-5.3-codex, gpt-5.6-sol and the gpt-5*-codex ids all
return 'not supported when using Codex with a ChatGPT account'; gpt-5.5 and
gpt-5.6-terra answer normally on the same account.

Closes #7521

changelog.d/fixes/7521-codex-test-probe-model.md

@@ -0,0 +1 @@
+- Fixed the Codex connection **Test** button always reporting success for ChatGPT-account tokens: the probe used `gpt-5.3-codex`, a codex-only model ChatGPT accounts reject with a 400 — the same status the probe treats as "auth OK", so a bad token was indistinguishable from a good one. It now probes with `gpt-5.5`, a model ChatGPT-account sessions actually support (#7521).

src/app/api/providers/[id]/test/route.ts

@@ -54,7 +54,12 @@ const OAUTH_TEST_CONFIG = {
       "User-Agent": "codex-cli/1.0.18 (macOS; arm64)",
     },
     // Minimal invalid body — triggers a fast 400 without consuming quota.
-    body: JSON.stringify({ model: "gpt-5.3-codex", input: [], stream: false, store: false }),
+    // #7521: probe with a ChatGPT-account-supported model. "gpt-5.3-codex" is a
+    // codex-only id that ChatGPT accounts reject with a 400 for the WRONG reason
+    // (unsupported model, not "auth ok, body invalid") — collapsing the auth signal
+    // so a bad token looks the same as a good one. "gpt-5.5" is served for
+    // ChatGPT sessions; `input: []` still yields the intended 400.
+    body: JSON.stringify({ model: "gpt-5.5", input: [], stream: false, store: false }),
     // 400 = bad request, but auth was accepted; only 401/403 means the token is bad.
     acceptStatuses: [400],
     refreshable: true,

tests/unit/oauth-connection-test-codex-endpoint.test.ts

@@ -67,6 +67,20 @@ test("codex test probes the real /responses endpoint and treats 400 as 'auth ok'
   const headers = (calls[0].init?.headers ?? {}) as Record<string, string>;
   assert.equal(headers.Authorization, "Bearer fake-codex-token");
   assert.ok(calls[0].init?.body, "must send a minimal body so the endpoint returns 400 (not 405)");
+
+  // #7521: the probe model must be one ChatGPT-account sessions actually support.
+  // "gpt-5.3-codex" is rejected outright for ChatGPT accounts ("The 'gpt-5.3-codex'
+  // model is not supported when using Codex with a ChatGPT account."), which also
+  // returns 400 — masking a bad token behind the same status code as a good one and
+  // making the Test button always report success. Assert the probe body carries a
+  // supported model instead.
+  const parsedBody = JSON.parse(String(calls[0].init?.body));
+  assert.equal(parsedBody.model, "gpt-5.5", "probe must use a ChatGPT-account-supported model");
+  assert.notEqual(
+    parsedBody.model,
+    "gpt-5.3-codex",
+    "probe must not use the codex-only model ChatGPT accounts reject (#7521)"
+  );
 });
 
 test("codex test reports invalid when the endpoint returns 401 (port PR#347)", async (t) => {