›_ Codex Issues
Codex Issues / #7732

Allow configurable reasoning efforts for Review Mode

Resolved 💬 2 comments Opened Dec 8, 2025 by voctory Closed Jan 31, 2026
💡 Likely answer: A maintainer (etraut-openai, contributor) responded on this thread — see the highlighted reply below.

What feature would you like to see?

The reasoning effort for the /review command was hardcoded to Low in this PR: https://github.com/openai/codex/pull/3701

From the PR alone, it's not clear why the review mode reasoning effort was set to Low. Proposing to allow a user-configurable reasoning effort, either by:

  1. Inheriting the current reasoning effort from the thread's config.model in codex-rs/core/src/codex.rs
  2. Creating a user preference for review mode reasoning effort in config.toml
  3. (if it's preferred to not allow configurability): At minimum, consider bumping the reasoning effort to Medium for more dynamic reasoning

From a local POC, it seems like approach 1 works well and creates much more comprehensive reviews than before (gpt-5.1-codex-max, xhigh effort inherited)

I'd be happy to raise a PR for this, seeking alignment first.

Additional information

_No response_

View original on GitHub ↗

2 Comments

rahul810050 · 8 months ago

Hii @voctory !!

Proposed Approach

I’d like to implement this by removing the hard-coded Low reasoning effort in the review-thread setup and letting /review inherit the current session’s model_reasoning_effort from Config.

Implementation outline:

  1. In codex-rs/core/src/codex.rs, inside the review-thread creation (spawn_review_thread),

delete the line forcing:

per_turn_config.model_reasoning_effort = Some(ReasoningEffortConfig::Low);
  1. Allow per_turn_config to keep whatever reasoning effort is already set in the session config (config.model_reasoning_effort).
  2. Keep model_reasoning_summary = Detailed (current intended review UX).
  3. Add a small test ensuring /review uses the inherited reasoning effort when present.

This approach maintains backward compatibility, avoids new config keys, and immediately enables higher-quality reviews for users who already configure higher reasoning efforts.

I’m happy to open a PR once confirmed.

etraut-openai contributor · 6 months ago

This feature request has not received enough upvotes in the past two months, so I'm going to close it.