Allow changing model/speed after a plan is generated before starting implementation
Codex app version
Desktop Codex: 26.602.40724 — released Jun 5, 2026
Platform
- macOS Tahoe
26.5.1 - Apple Silicon MacBook Pro
- Pro plan
Summary
Please improve the UI so users can change the model/speed selection after Codex generates a plan but before starting implementation.
A common workflow is to use a faster/cheaper mode to generate the plan, then switch to a slower/stronger mode before implementing the task. For example, I often use GPT-5.5 Fast to write the plan, and then switch to GPT-5.5 Standard to execute the implementation when I am not in a hurry.
However, after Codex presents the plan and asks whether to implement it immediately, the confirmation panel blocks or covers the normal model/speed selector area. This makes it inconvenient or impossible to switch the model/speed before pressing the implementation button.
Observed behavior
- Ask Codex to create a plan.
- Codex generates the plan.
- Codex displays an
Implement this plan?confirmation panel with options such as:
Yes, implement this plan
No, and tell Codex what to do differently- This confirmation panel visually covers the bottom area where the model/speed selector normally appears.
- The user cannot easily switch from the planning mode/model to the desired implementation mode/model before starting the task.
Expected behavior
The model/speed selector should remain accessible after the plan is generated and before implementation begins.
Users should be able to:
- generate the plan with one model/speed,
- review the plan,
- switch to a different model/speed for execution,
- then press
Yes, implement this plan.
Actual behavior
The implementation confirmation UI can block the model/speed selector, making it hard to switch the model or speed before starting implementation.
Why this matters
Planning and implementation often benefit from different model/speed choices. A fast mode can be good enough for quickly drafting a plan, while a stronger/slower mode may be preferred for actually editing files, running tests, and making final decisions.
Blocking the model/speed switcher at exactly the moment before implementation reduces workflow flexibility and can cause users to start implementation with the wrong mode.
Request
Please adjust the plan confirmation UI so it does not cover the model/speed selection controls. Possible solutions:
- keep the model/speed selector visible above or inside the plan confirmation panel,
- provide a model/speed dropdown directly in the
Implement this plan?panel, - allow the user to change execution mode before confirming implementation,
- or make the confirmation panel layout avoid covering the normal composer controls.
This would make the plan-then-execute workflow much smoother for users who intentionally use different model/speed settings for planning versus implementation.
3 Comments
This is exactly the interaction I want. Rather than adding another dropdown inside the “Implement this plan?” panel, keep the existing model control available as the trigger for an expandable routing tray beneath the composer.
The tray would provide direct, parallel controls for:
model: Luna · Terra · Sol · Spark
reasoning effort: Low · Medium · High · Ultra
speed: ⚡ Fast toggle
Each should be a segmented, one-click control—not a menu. It could be opened to change the route, then dismissed again, or left open while iterating. Crucially, the plan-confirmation panel must not cover it.
That would solve the immediate planning-to-implementation hand-off in this issue, while also making routine in-thread model, effort and speed changes far less cumbersome. I’ve attached a quick concept mock-up.
<img width="1672" height="941" alt="Image" src="https://github.com/user-attachments/assets/4b46f016-6e88-4af9-a1de-18199a55fb4a" />
Please allow model selection/switching directly from the “Implement this plan?” action in the Codex app. I often want to use a stronger model or higher reasoning effort for planning, then switch to a faster or more cost-efficient model for implementation. The current workaround—dismiss the prompt, change the model, and manually ask Codex to implement—adds friction and is easy to miss. A model/reasoning selector in the implementation confirmation would make this workflow much clearer and more efficient.
看完 wiki 回来，干货满满，https://520menghuan.pages.dev/ 的日漫站搭建教程要是能配合这份文档就完美了。