[Bug]: Custom models can be added to ChatGPT-connected Codex instances, but the server excludes them #16004

Before submitting

• I searched existing issues and did not find a duplicate.
• I included enough detail to reproduce or investigate the problem.

Area

apps/web

Steps to reproduce

1. In Settings, add a Codex instance using "Connect with ChatGPT" and finish signing in.
2. Open that instance's settings and go to the Models section.
3. Add a custom model using a slug that isn't already in the list. If you enter a slug that's already there, it's rejected as a built-in.
4. Start a new chat and pick the custom model in the model picker.
5. Look at the composer controls: there's no Reasoning control.
6. Switch to a built-in model on the same instance. The Reasoning control appears.

Expected behavior

Custom models should behave the same way in the settings UI and on the server for ChatGPT-connected Codex instances. Either the Models settings shouldn't offer custom models for this setup mode, so they can't appear in the picker without capabilities, or the server should keep them in the instance's model list, as it does for CLI-mode instances, so they get a Reasoning control. I'm not sure which is intended, since the managed driver seems to deliberately limit the list to the account's OpenAI catalog.

Actual behavior

On a Codex instance connected with "Connect with ChatGPT," a custom model added in Settings appears in the model picker, but selecting it shows no Reasoning control, while built-in models on the same instance show one

Impact

Minor bug or occasional failure

Version or commit

No response

Environment

macOS, desktop app version 0.0.45

Logs or stack traces

Screenshots, recordings, or supporting files

No response

Workaround

No response