Claude Code Review is Anthropic's managed pull request reviewer. When a PR opens on GitHub, it sends several Claude agents to check the changes in parallel, verifies what they find, and posts ranked comments back on the PR. It was launched on 9 March 2026 as a research preview for Claude Team and Enterprise customers, and a typical review costs $15 to $25. This guide covers how it works, what it costs, how to set it up, and when a cheaper tool makes more sense.
Claude Code Review at a glance
What it does: Reviews GitHub pull requests with multiple agents running in parallel.
Who can use it: Claude Team and Enterprise customers (research preview at launch).
Cost: About $15 to $25 per review, billed on token usage, separate from your seat price.
Speed: Around 20 minutes per review on average.
Setup: An admin enables it and installs the Anthropic GitHub app.
Customization: REVIEW.md for review rules, CLAUDE.md for project context.
Things may have changed since launch, so check Anthropic's Claude Code documentation for current plans and limits before you commit.
What is Claude Code Review?
Claude Code is Anthropic's agentic coding tool. Code Review is a separate, managed feature built on top of it. Instead of you asking Claude to look at a difference, the review runs on its own whenever a pull request opens in a repository your admin has switched on.
The point is volume. Teams that use AI to write code now open more pull requests than humans can carefully review. Anthropic built Code Review to catch the bugs so that human reviewers can spend their time on design and intent.
How does Claude Code Review work?
When a PR opens, Claude dispatches a team of agents. They are not sequential. Each takes a different look at the change. Then they pass through a verification process that strives to eliminate false alarms. What remains after the verification is then sorted by severity and added as comments to the pull request.
In practice, there are two things to consider. The review examines both differentiated and non-differentiated code. It can expose a bug that doesn't appear until the change is applied to code elsewhere. The numbers are what make Anthropic so confident about its low error rate, and the verification pass accounts for that.
What are the outcomes that Anthropic is reporting?
These statistics are based on the facts and numbers that Anthropic uses internally and should not be considered an independent test.
Before Code Review, 16% of pull requests had substantive review comments.
Engineers marked fewer than 1% of the findings as incorrect.
On large PRs of more than 1,000 changed lines, 84% got findings, averaging about 7.5 issues each. On PRs under 50 lines, 31% got findings, averaging about half an issue.
The size pattern is the useful part. Big changes get the most out of it. Tiny changes often don't need it.
How much does Claude Code Review cost?
Reviews are billed on token usage, and Anthropic puts the typical cost at $15 to $25 per pull request. Price rises with the size and complexity of the change. It's billed separately from the usage included in a Team or Enterprise seat, and there's no free tier.
Here's what that looks like at a glance:
Reviews per week | At $15 each | At $25 each |
10 | $150 | $250 |
20 | $300 | $500 |
50 | $750 | $1,250 |
It's easy to do the maths, but it surprises teams. When each little PR receives a thorough review, the bill goes up quickly. Anthropic provides admins control over this: a monthly spending limit for the organisation, an on/off toggle per repository, and an analytics dashboard that monitors review counts, acceptance rates and cost.
Who can use it?
At launch, the managed GitHub service was limited to Team and enterprise organisations. It was not available for organisations with Zero Data Retention turned on.
Sources disagree about individual plans. Some say Pro and Max users can run a /code-review command inside Claude Code, while the managed GitHub service stays Team and Enterprise only. Confirm this in Anthropic's docs before you plan around it.
How to set it up
A Claude Team or Enterprise admin enables Code Review in the Claude Code settings.
The admin installs the Anthropic GitHub app on the organisation.
Choose the repositories that should get automatic reviews.
Optionally add a REVIEW.md file to describe what the reviewer should care about and a CLAUDE.md file to give it project context.
Set a monthly spending cap before you turn on busy repositories.
Start with one or two active repos, watch the cost and the comment quality for a couple of weeks, then widen it.
Claude Code Review vs the alternatives
There are cheaper ways to get AI help on pull requests, and they work differently.
Particulars | Claude Code Review | Claude Code GitHub Action | GitHub Copilot review |
Style | Managed, multi-agent, with verification | Open source, lighter checks | Single-pass review |
Cost | About $15–25 per review | Your own API usage | Uses a Copilot premium request per review |
Best for | Large or risky changes | Everyday light checks | Teams already on Copilot |
A few notes. The open-source GitHub Action stays available and costs far less, but it's less thorough by design. Copilot's review is bundled with its subscription allowance, so it's usually cheaper per PR, though it makes a single pass instead of a multi-agent one. Another option is a dedicated review product like CodeRabbit. Try to compare them on your own pull requests, as it really depends on your codebase.
If budget is an issue, a sensible configuration is to have a light tool for all PR and Claude Code Review for large or security-sensitive PRs.
What are the restrictions?
Price – If the repository is large, it will cost $15 to $25 per review, and there should be a limit and a budget.
Availability – Associations with terms and features may change; it began as a research preview and is tied to Team and Enterprise plans.
Time- Most PRs will take about 20 minutes for review, whereas a quick fix will probably take under five minutes.
Data rules – It could not be used by Zero Data Retention organisations at launch.
It's still AI—it may not understand context or mark irrelevant items. There has to be somebody who reads the comments.
No approvals – It leaves comments. It doesn't merge anything or give it the thumbs up.
Is Claude Code Review worth it?
Depending on how much a bug costs you. A review that finds one serious bug could be worth the effort for teams that ship big changes, have sensitive code, or are deluged with AI-generated pull requests. For a small team joining in small PRs, the lighter options will probably suffice.
Final verdict
Claude Code Review is a serious tool aimed at teams with serious review backlogs. The multi-agent design and the verification step are what set it apart, and Anthropic's internal numbers are strong, though they come from Anthropic itself. The price keeps it from being a default for every PR. Pilot it on a couple of busy repositories, set a spending cap, and compare it against a cheaper option on your own code before you roll it out.
Want help setting up AI code review or building AI agents into your workflow? Our AI Development team builds them, and our AI Consulting team can help you choose the right tools first. Book a free consultation.
For more on where AI is heading, read The Future of Generative AI: 8 Key Trends.
Frequently asked questions
What is Claude Code Review?
It's Anthropic's managed pull request reviewer. It uses several Claude agents in parallel to check changes on GitHub, verifies the findings and posts ranked comments.
How much does Claude Code Review cost?
Typically $15 to $25 per review, billed on token usage and separate from your seat price. The final cost depends on the size and complexity of the PR.
Is Claude Code Review free?
No. There's no free tier. It requires a Team or Enterprise plan, and review usage is billed on top.
Who can use Claude Code Review?
Claude Team and Enterprise customers, as a research preview at launch. Check Anthropic's docs for current availability, including rules for Zero Data Retention organizations.
How long does a review take?
About 20 minutes on average, depending on the size of the change.
Does it replace human code reviewers?
No. It doesn't approve pull requests. It surfaces issues so human reviewers can focus on design and intent.
How do I customize what it checks?
Add a REVIEW.md file with your review criteria and a CLAUDE.md file with project context.
Is there a cheaper option?
Yes. Anthropic's open-source Claude Code GitHub Action is lighter and costs less, and GitHub Copilot offers its own review feature. Many teams use a light tool broadly and the full multi-agent review for big changes.