Feature request: Model + Reasoning presets/profiles

What variant of Codex are you using?

App

What feature would you like to see?

Problem

Codex lets users choose a model and a reasoning effort, but these settings have to be changed independently.

For users who regularly switch between different levels of complexity, this creates unnecessary friction. In practice, model + reasoning effort are often treated as a single decision, for example:

• Fast / simple: Luna + Low
• Default: Terra + Medium
• Deep troubleshooting: Sol + High

For example, when moving from a straightforward question to a difficult debugging task, I may want to switch from:

Terra + Medium → Sol + High

Today, this requires manually changing two separate settings.

Proposed solution

Allow users to create presets/profiles that combine a model and reasoning effort, and provide a fast way to switch between them.

For example:

Default       → Terra + Medium
Deep          → Sol + High
Fast          → Luna + Low

Ideally, these could be assigned keyboard shortcuts:

Ctrl/Cmd+1 → Default
Ctrl/Cmd+2 → Deep
Ctrl/Cmd+3 → Fast

Alternatively, a preset menu such as:

Fast       Luna · Low
Default    Terra · Medium
Deep       Sol · High

would already solve most of the problem.

Possible implementation

A minimal version could simply support saved named combinations of:

• Model
• Reasoning effort

For example:

{
  "name": "Deep",
  "model": "Sol",
  "reasoning": "high"
}

Users could then select the preset from the model/reasoning UI or assign an optional keyboard shortcut.

This would also leave the existing individual controls intact for users who want fine-grained control.

Expected benefit

This would make Codex much faster to operate for users who frequently switch between:

• routine technical questions
• normal development work
• difficult debugging / troubleshooting

It seems like a relatively small UX feature that could have a substantial usability impact.

Thanks!

Additional information

No response

Start by locating the App UI code that controls model selection and reasoning effort. Use the proposed named combinations such as Default, Deep, and Fast as the behavior to evaluate; done means users can save and switch a model-plus-reasoning preset while retaining the existing individual controls. No file or test is named in the issue.