Codex Desktop ignores `plan_mode_reasoning_effort` when entering Plan mode, while CLI applies it correctly

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

Not captured at the time of filing.

What subscription do you have?

Not captured at the time of filing.

What platform is your computer?

macOS

What issue are you seeing?

Codex Desktop does not apply plan_mode_reasoning_effort when switching a thread into Plan mode.

My config contains:

model_reasoning_effort = "medium"
plan_mode_reasoning_effort = "xhigh"

With this setup:

• In Codex CLI, entering Plan mode correctly switches the reasoning effort from the global/default value to the Plan-mode override (xhigh).
• In Codex Desktop, entering Plan mode does not switch the reasoning effort accordingly.
• I can visually confirm in the Desktop UI that the reasoning level remains medium after entering Plan mode, instead of switching to xhigh.

So the same configuration works in CLI but not in the Desktop app.

What steps can reproduce the bug?

1. Set the following in ~/.codex/config.toml:

model_reasoning_effort = "medium"
plan_mode_reasoning_effort = "xhigh"

1. Open Codex CLI and enter Plan mode.
2. Observe that Plan mode uses the configured override (xhigh).
3. Open Codex Desktop with the same config.
4. Enter Plan mode in a thread.
5. Observe in the Desktop UI that the reasoning level remains medium instead of switching to xhigh.

What is the expected behavior?

When plan_mode_reasoning_effort is explicitly set, Codex Desktop should apply that value on entering Plan mode, the same way Codex CLI does.

In this example, entering Plan mode should switch reasoning effort from medium to xhigh, and the Desktop UI should reflect that change.

Additional information

This looks like a Desktop-specific config/application bug rather than a request for a new feature, because the config key already exists and the CLI respects it.

The issue is specifically that Desktop and CLI behave differently for the same configuration.

If useful, I can also provide:

• exact Codex Desktop version
• exact macOS version / architecture
• screenshot or screen recording showing the reasoning level before and after entering Plan mode
• session ID from in-app feedback