Codex App: /review silently downgrades GPT-5.6 Sol from Ultra to Max

What version of the Codex App are you using (From “About Codex” dialog)?

Codex App 26.707.51957 (Build 5175).

What subscription do you have?

ChatGPT Pro 20×.

What platform is your computer?

macOS 26.5.2 (Build 25F84), arm64.

What issue are you seeing?

In the Codex desktop app, I explicitly select GPT-5.6 Sol with Ultra intelligence and then start the built-in /review workflow. As soon as the review starts, the selected intelligence level immediately changes from Ultra to Max.

This appears to affect execution, not only the displayed label: the subagents that normally start automatically with Ultra are also missing.

I reproduced the behavior twice:

• once after the app had been open for a long time;
• once again after fully closing and restarting the app.

If /review intentionally does not support Ultra, the app should communicate that limitation before starting. It should not silently change an explicitly selected setting.

What steps can reproduce the bug?

1. Open a repository in the Codex desktop app.
2. Select GPT-5.6 Sol.
3. Select Ultra intelligence.
4. Start the built-in /review workflow.
5. Observe that the UI immediately switches the intelligence level from Ultra to Max.
6. Observe that the Ultra subagent/delegation behavior is absent.

Feedback ID: 019f58b8-0c66-7f83-bdba-22d3e6d9861e

What is the expected behavior?

The /review workflow should preserve and run with the explicitly selected Ultra setting, including Ultra's normal subagent behavior.

If review mode has an intentional Max-only limitation, the app should explain it before starting and require confirmation rather than silently changing the selected intelligence level.

Additional information

This is distinct from #30585, which reports the wire-level mapping of Ultra to reasoning.effort=max being rejected by a backend. Here, the Codex App visibly changes the user-facing selection from Ultra to Max when /review starts, and the expected Ultra delegation behavior is missing.

Related: #32084 mentions a review being performed by GPT-5.6 Sol Max inside a broader multi-bug report, but does not isolate or document this reproducible Ultra-to-Max transition.