Automated code review: Copilot, Claude Code and Codex compared

Automated code review means asking a model to read a pull request and comment on bugs before a human approves it. On October 23, 2026 the three tools most teams compare are GitHub Copilot code review, Claude Code's /code-review and Codex code review. They do the same job in different places: Copilot inside GitHub's reviewer list, Claude Code in your terminal or as a deeper cloud review, and Codex in GitHub through a comment or automatic reviews. This guide compares how each one works, what it costs and how to make it useful, based on the official documentation read on October 23, 2026.

The three tools at a glance

Table: code review tools (official documentation, checked October 23, 2026)

GitHub Copilot code review

On a pull request, click Request next to Copilot under Reviewers. GitHub says the review usually takes less than 30 seconds. Copilot labels each comment High, Medium or Low, includes suggested changes you can apply in a couple of clicks, and adds an approval assessment to its overview comment. By default it leaves a Comment review, which does not count toward required approvals; administrators can allow it to approve, and that approval is dismissed if new commits arrive.

Automatic reviews are available on Copilot Pro, Pro+ and Max and with Business or Enterprise licenses; with a ruleset you can also re-review every new push. Copilot Free does not include code review. Two limits worth knowing: Copilot does not see replies you leave on its comments, and on a re-review it may repeat comments you already resolved or downvoted.

Claude Code /code-review

In Claude Code, /code-review reviews the changes on your current branch, including uncommitted ones, and reports correctness bugs in the conversation. You can point it at a pull request number, a branch, a path or a commit range, for example /code-review 1234 or /code-review main...my-feature. It runs as a background subagent with its own context, so it does not fill your session, and it follows your CLAUDE.md. Two flags change what happens: --fix applies the findings to your working tree, through your permission mode, and --comment posts them on a GitHub pull request as inline comments. The effort level trades coverage for confidence: low reports only the findings it is most sure of.

It works on any plan and counts toward your usual usage. For important merges there is also /code-review ultra, a research preview that runs a multi-agent review in the cloud and reproduces each finding before reporting it. Pro and Max accounts get three free runs, once; after that Anthropic says a review typically costs $5 to $25 in usage credits, depending on the size of the change. Our guide on how to use Claude Code covers the rest of the tool.

Codex code review

Codex reviews GitHub pull requests in two ways: mention @codex review in a comment, or connect the repository, open Codex code review settings and turn on Automatic review. Codex posts its review like a teammate would and, in GitHub, flags only P0 and P1 issues to keep the comments on high-priority risks. It follows the AGENTS.md files in your repository: add a ## Code Review Rules section to the file closest to the code those rules govern. Our guide to AGENTS.md explains the file. GitLab merge request support is in preview.

How to make reviews worth reading

1. Write the rules down. All three read an instructions file: tell the reviewer what matters in your codebase, such as security checks, error handling or performance budgets, and what to ignore.
2. Keep pull requests small. Every tool reviews a diff; smaller diffs give sharper findings and cheaper cloud reviews.
3. Treat findings as suggestions. A model can be wrong, and a confident wrong comment is still wrong. Reproduce a bug before you fix it.
4. Do not let it approve alone. Copilot can approve when allowed, but a human should still own the merge for anything that matters.
5. Watch what it can reach. Reviews that read issues, pages or connected tools can be steered by prompt injection hidden in that content.

Which one to pick depends mostly on where your team already works: Copilot if you live in GitHub with Copilot licenses, Codex if you use ChatGPT and want only high-priority comments, and Claude Code if you want reviews in the terminal before you even open the pull request. Our comparison of the best coding assistants covers the tools behind them.