Codex GitHub code review usage increased significantly on ChatGPT Business — can I lower the model or control review token usage?

Codex Codex CLI
1

Hi everyone,

I’m using ChatGPT Business and have connected Codex Cloud to GitHub through the Codex GitHub connector. My usage is almost exclusively for code reviews, not for feature implementation or general coding tasks.

My workflow is:

  1. Open a new PR.
  2. Ask Codex to review the PR.
  3. If Codex finds P0–P2 issues, I move the PR back to draft.
  4. I fix the issues.
  5. I reopen the PR and ask Codex to review it again.
  6. I repeat this only when needed.

I do not use Codex heavily for writing new code. It is mainly used as a PR review assistant to catch high-priority issues before merging.

However, starting around two weeks ago, the usage/token consumption rate feels much faster than before, even though my workflow has not changed much. I am trying to understand whether something changed in how Codex GitHub code reviews consume credits/tokens, or whether my repository/context/diff size is causing unexpectedly high usage.

A few questions:

  • Is there any way to choose or lower the model used by Codex GitHub code review on ChatGPT Business?
  • Can I configure Codex GitHub reviews to use a lighter model, such as a mini model, for routine PR reviews?
  • Is the GitHub PR review model fixed by OpenAI, or can it be controlled through repository settings, Codex settings, AGENTS.md, or Business workspace settings?
  • Are there recommended ways to reduce token usage for Codex GitHub code reviews?
  • Does reopening a PR after moving it from draft cause Codex to review the full PR context again, rather than only the new changes?
  • Are automatic reviews and manual @codex review charged differently, or do they use the same review pipeline?

My ideal setup would be:

  • Use a lighter/cheaper model for routine PR reviews.
  • Reserve the stronger model only for security-sensitive or complex backend changes.
  • Limit reviews to P0/P1/P2 issues only.
  • Avoid spending a large amount of usage on style, formatting, naming, or minor maintainability comments.

Has anyone else on ChatGPT Business noticed a significant increase in Codex GitHub code review usage recently? And is there currently any supported way to control the model or reduce review token usage?

Thanks.

1 Like