Regression: split model and intelligence selector adds too many clicks

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

Codex App 26.422.21637 (2056)
VScode extension v26.5422.21459 (pre-release)

What subscription do you have?

ChatGPT Pro

What platform is your computer?

Darwin 25.3.0 arm64

What issue are you seeing?

The rebuilt model selector in the latest Codex App release is a UX regression for frequent model switching.

I switch between model/intelligence pairs depending on the task, most commonly:

• gpt-5.5 + medium
• gpt-5.3-codex + high

With the new split model and intelligence controls, switching between these pairs requires an awful lot of clicks. It interrupts the workflow every time I move between a cheaper/faster mode and a deeper coding mode.

The previous model/reasoning selector flow was much faster and worked well. Please restore that previous behavior, or provide an equivalent one-action path for switching common model/reasoning pairs.

One possible fix would be to attach remembered intelligence/reasoning settings per model. For example, if I last used gpt-5.5 with medium, then selecting gpt-5.5 should restore medium; if I last used gpt-5.3-codex with high, selecting gpt-5.3-codex should restore high. That would make model switching do the full task in one click.

This is separate from the current bugs where the selector is stuck or does not persist. The issue here is the extra interaction cost from splitting model and intelligence selection in the new UI.

What steps can reproduce the bug?

1. Open Codex App 26.422.21637.
2. Start or open a chat.
3. Select gpt-5.5 with medium intelligence.
4. Later switch to gpt-5.3-codex with high intelligence.
5. Switch back again to gpt-5.5 with medium.
6. Repeat during normal work.

Observed behavior:

• Model and intelligence must be adjusted separately.
• Common model/intelligence pairs are not restored as a unit.
• Frequent switching requires multiple clicks and breaks focus.

What is the expected behavior?

Switching between common model/intelligence pairs should be possible in one action, as it effectively was before.

Expected behavior could be either:

• Restore the previous fast model/reasoning selector UX, or
• Remember intelligence per model and apply the remembered intelligence automatically when selecting a model (maybe that was the case, already - never had to switch medium/high)