codex - 💡(How to fix) Fix Feature request: add `plan_mode_model` config override for Plan mode [1 comments, 2 participants]

What variant of Codex are you using?

CLI / TUI Plan mode

What feature would you like to see?

Add a first-class plan_mode_model configuration option that lets users choose the default model used by Plan mode independently from the global model.

Today, Plan mode follows the built-in Plan preset model, which currently inherits the global model. That means users who want one model for planning and another model for normal execution need to manually switch models before/after using Plan mode, or keep separate profiles/sessions.

Requested behavior:

• Support a top-level config value:

  model = "gpt-execute"
  plan_mode_model = "gpt-plan"

• Support profile-level overrides:

  plan_mode_model = "gpt-top-plan"
  profile = "dev"
  
  [profiles.dev]
  plan_mode_model = "gpt-profile-plan"

  When profile dev is active, Plan mode should use gpt-profile-plan.

• If plan_mode_model is unset, preserve the existing behavior: Plan mode uses the built-in Plan preset model, which currently follows the global model.

• When switching into Plan mode, the TUI should apply the configured Plan-mode model override.

• When submitting a message with Plan mode active, the submitted collaboration mode should use the configured Plan-mode model.

• When the model/reasoning selector is used while Plan mode is active, users should be able to choose whether the change applies only to the Plan-mode override or to both the global default and the Plan-mode override.

Why this matters

This makes Plan mode configurable in the same spirit as other mode/phase-specific settings.

It also makes the Plan-mode workflow less error-prone:

• users no longer need to manually switch models before entering Plan mode;
• the execution model can remain stable while Plan mode uses a different default;
• profile-specific defaults make it possible to have different Plan-mode model choices per team, project, or workflow;
• clearing the override cleanly returns Plan mode to the built-in preset behavior.

Example configuration

model = "gpt-5.4"

plan_mode_model = "gpt-5.2"
plan_mode_reasoning_effort = "high"

Resulting behavior:

• normal execution uses gpt-5.4;
• Plan mode uses gpt-5.2;
• Plan mode uses the configured Plan-mode reasoning effort;
• removing plan_mode_model restores the existing Plan preset/global-model behavior.

Acceptance criteria

• plan_mode_model is accepted in config.toml.
• plan_mode_model is accepted under [profiles.<name>].
• Profile-level plan_mode_model overrides the top-level value.
• The generated config schema includes plan_mode_model.
• The TUI applies the Plan-mode model override when Plan mode is active.
• The TUI persists Plan-mode-only model changes to the appropriate config path:
  • root path when no profile is active;
  • [profiles.<active_profile>] when a profile is active.
• Clearing plan_mode_model resets Plan mode to the built-in Plan preset model.
• Documentation explains the new config option and fallback behavior.

Non-goals

• This does not change the global model behavior.
• This does not introduce automatic model routing.
• This does not change /review model behavior.
• This does not require Plan mode to always use a different model; it only adds an optional override.

Additional information

Related requests:

• #13206 asks for a dedicated model config for /plan.
• #5558 asks for separate model selection for Plan vs Execute modes.
• #18406 asks for better model switching inside the Plan-mode approval flow.

This request is narrower: add a persisted, profile-aware Plan-mode model default via plan_mode_model, while preserving the current fallback behavior when the setting is absent.

Example implementation

A prototype implementation is available in https://github.com/aaudin90/codex/pull/1.

https://github.com/user-attachments/assets/f76c7ebb-7437-4b52-ac40-614764403fba