Claude Code Review leaves preview: PR reviews 3x faster

On October 15, 2026, Anthropic announced that Claude Code Review is leaving research preview and is now available to organizations on the Claude Team and Enterprise plans. It’s a service in which several Claude agents review every pull request on GitHub and leave comments right on the lines of code.

What’s new in Claude Code Review?

According to the October 15 announcement, three things changed:

1. Reviews are roughly 3x faster: instead of about 20 minutes, they now take 6–10 minutes.
2. Comments are meant to catch more of the bugs that actually matter, and they’re shorter: the problem, what causes it and how to fix it.
3. An admin can set a Quick, Standard or Deep level for each repository. Standard is the default, Quick is meant for routine projects, and Deep for the most business-critical ones.

On top of that, eligible Enterprise organizations can claim at least $10,000 in Code Review credits until January 15, 2027 (the credits expire 30 days after you claim them).

How does Claude Code Review work?

The documentation describes a review in four steps, run on Anthropic’s infrastructure:

1. Finding: several agents read the change in parallel, together with the surrounding code.
2. Verification: a separate step checks the problems found against the code and filters out false positives.
3. Ranking: whatever is left gets deduplicated and sorted by severity.
4. Publishing: the comments land in the PR as inline comments, with a summary and a check run named “Claude Code Review”.

Obviously, a review never approves or blocks a PR, so the merge decision stays with humans. By default it focuses on correctness (bugs that would break production), not formatting. The code is only read: the review doesn’t build it, run it or push commits.

We can tune the reviewer’s behavior with REVIEW.md and CLAUDE.md files in the repository. A security curiosity: REVIEW.md is always read from the base branch, never from the changes in the PR itself, so there’s no way to “smuggle” instructions into the code under review ;)

How does Anthropic use it internally?

The same day, the Claude Code team wrote on its blog about how it does code review itself, now that Claude writes about 90% of Claude Code’s code. A few numbers from that post:

1. The number of merged PRs in Claude Code more than doubled from June to September, while the share of reverted changes dropped by over 40%.
2. Across Anthropic’s whole monorepo, the number of changes per day grew from about 2,500 in May to over 30,000 in September.
3. The share of PRs that pass after one round of review and fixes rose from 65% to 78%, and the share needing five or more rounds fell from 8.9% to 1.5%.

The team hasn’t (yet) given up on humans. It sorts changes by risk: small changes behind a flag don’t need human approval once tests are green, regular changes need one person, and risky ones need two.

Plus one simple rule: every reviewer comment must get a response, either a fix or a closed thread with a justification.

How do you turn on Claude Code Review?

According to the announcement, setup is the same as in the preview:

1. An organization Owner enables Code Review in the Claude Code settings.
2. They install the Claude app on GitHub and pick the repositories.
3. For each repository, they set when reviews should run: when a PR is opened, after every push, or only on request.

After that, reviews run on their own, and anyone with write access can trigger another one by commenting @claude review. The service isn’t available to organizations with Zero Data Retention, a HIPAA configuration or customer-managed encryption keys (CMEK).

Without a Team or Enterprise plan we can still try something: the /code-review command runs a lighter review locally, in a Claude Code session, on any plan. If we’re just getting started, the Claude Code tutorial from scratch will come in handy.

FAQ

How much does Claude Code Review cost?

The announcement doesn’t give pricing, apart from the Enterprise credits. According to the documentation, the cost grows with the review level: Quick is the cheapest, Deep the most expensive, and an admin can set a monthly spending limit.

Does Claude Code Review approve pull requests?

No. It only publishes comments and a check run that doesn’t fail. A human always approves the merge.

Does Code Review work with GitLab?

The service works with repositories on GitHub, including GitHub Enterprise Server. On GitLab you can run Claude in your own CI.

How is Code Review different from /code-review?

Code Review is a managed service for Team and Enterprise that runs automatically on PRs. /code-review is a command in Claude Code that we run ourselves on our own branch, on any plan.