Claude Code Review is generally available for Team and Enterprise, with Quick, Standard, and Deep levels. Here's the setup, real pricing, and the EU credit catch.
Vamsi Tallapudi
Manager, Architect Technology at Cognizant
Claude Code Review left research preview on October 15, 2026. It's now generally available for every Claude Team and Enterprise organization, reviews finish in 6 to 10 minutes instead of the ~20 minutes it took during preview, and admins can now pick a review depth of Quick, Standard, or Deep for each repo. That's the news. Whether you should flip it on for your repos is a different question, and the answer depends more on your git history than on anything Anthropic put in the announcement.
I've been digging through the launch post, the pricing fine print, and the Hacker News thread arguing about it, so let's get into what actually changed and where the catch is.
Code Review is Anthropic's automated pull-request reviewer: a team of Claude agents reads your diff, checks it against the rest of the codebase, and leaves comments on GitHub like a human reviewer would. It's been in research preview since earlier this year. GA mostly means three things changed.
First, speed. Anthropic says most reviews now finish about 3x faster than during preview, dropping from roughly 20 minutes down to 6 to 10 minutes. For a team that's used to a reviewer commenting before lunch, that's the difference between "useful" and "actually in the loop."
Second, noise. The preview version apparently flagged a lot of stuff nobody cared about: formatting nits, missing test coverage, style preferences. The GA version is tuned to focus on correctness, bugs that would actually break production, and each finding is supposed to state the problem, what triggers it, and a fix, in a few sentences instead of a wall of text.
Third, and this is the part worth paying attention to: admins now set review depth per repository instead of getting one setting for everything.
Here's the breakdown as Anthropic describes it:
| Level | What it's for | Rough trade-off |
|---|---|---|
| Quick | Routine repos, fast iteration | Fewer agent passes, lowest cost, catches the obvious stuff |
| Standard | The default for most repos | Balanced coverage, this is what you get if you don't touch the setting |
| Deep | Critical repos where a missed bug is expensive | More agent resources per PR, thorough, costs more and takes longer |
The logic is straightforward once you say it out loud: your internal tooling repo doesn't need the same scrutiny as your payments service. Set Quick on the low-stakes stuff and save Deep for the repos where a bad merge means an incident, not an annoyance. That per-repo control didn't exist in preview, where you got one review intensity across your whole org whether you wanted it or not.
This is genuinely simple, which is rare for an enterprise AI feature. For admins:
That's it. Developers don't configure anything. Once it's on, every new pull request to an enabled repo gets a review automatically. No new workflow to learn, no slash command to remember.
Here's where it gets murkier, because Anthropic's own posts are light on hard numbers and most of what's online comes from people who've tested it independently rather than Anthropic's pricing page. What's confirmed: it's billed on token usage, not a flat per-seat fee, and it's only available to Team and Enterprise plans, not Pro or Max. Admins get monthly organization spend caps, repo-level toggles, and an analytics dashboard showing PRs reviewed, acceptance rate, and total spend.
During the preview period, independent testing put typical reviews around $15 to $25 each, scaling up with PR size, with reports of small PRs costing single digits and very large, history-heavy PRs running $30 to $40. Take those as ballpark figures from third-party tests, not an official rate card, since Anthropic hasn't published per-review pricing for the GA version.
What Anthropic has published clearly: eligible Enterprise customers can claim $10,000 to $50,000 in Code Review credits, with the exact amount set by active Claude Code seat count at the moment you claim. The window closes January 15, 2027, and once claimed, credits expire after 30 days. So don't claim early and sit on them.
This is the detail that got buried under the "$50,000 in free credits" headline, and it matters if you're running anything regulated out of Europe. Organizations on GitHub Enterprise Cloud with data residency, the .ghe.com setup most EU firms use specifically to keep code inside the EU, cannot claim these credits at all. Worse: if you claim credits first and then connect a data-residency GitHub instance afterward, you lose the credits for your whole org, including reviews running on plain github.com. Firms running zero data retention can't use Code Review at all, full stop.
The product itself still works for EU teams on data residency setups. They just don't get the free money. If your compliance team picked .ghe.com for a reason, read that fine print before you tell finance you've got $50k coming.
The number Anthropic leans on hardest: 54% of pull requests now get substantive comments, up from a 16% baseline, with less than 1% of findings reported as wrong. Internally, Anthropic says merged PRs roughly doubled between June and September, reverts dropped more than 40%, and the average number of review rounds needed to pass fell from 2.08 to 1.32. First-round pass rate went from 65% to 78%.
Those are real numbers, but notice whose homework is getting graded: Anthropic is reporting on Anthropic's own repos, where Claude reportedly writes about 90% of Claude Code's code in the first place. That's the company checking whether its reviewer catches mistakes made by its own coding agent. I'd want to see those numbers from a team outside Anthropic before treating them as gospel, and so far nobody's published an independent replication.
The Hacker News thread discussing this is more skeptical than the press releases, and the skepticism is worth sitting with. One commenter's line stuck with me: relying on an AI reviewer to explain a system you didn't build yourself is like being "the owner of a recovered alien spacecraft with tech you have no hope of understanding." Another pointed out that more PRs merging faster isn't automatically a win if nobody on the team can explain how the thing works anymore. That's not a knock on Code Review specifically, it's the general tension with any tool that makes merging easier without making understanding easier.
There's also a competitive tell worth noting: CodeRabbit's CEO recently said OpenAI models are now his primary supplier for their reviewer, a shift from Anthropic six months earlier. If a company that sells AI code review for a living is hedging away from Claude models, that's a data point, not proof of anything, but it's not nothing either.
Honestly, yes, if you're already paying for Claude Team or Enterprise seats. It's close to free to try: install the GitHub App, set a couple of low-stakes repos to Quick, and watch what it flags for two weeks before deciding whether to pay for Deep on anything that matters. The setup cost is basically zero, which is the strongest argument here.
Where I'd slow down: don't let "Code Review passed" become a substitute for a human actually reading the diff on anything that touches money, auth, or data. The acceptance-rate numbers are good, but they're self-reported on a codebase where Claude wrote most of the code being reviewed. And if you're an EU company that picked GitHub data residency for a real compliance reason, don't plan your budget around that $50k before checking whether you actually qualify.
If you're not already in the Claude ecosystem, this isn't a reason to switch. It's a nice perk for people already there, not a standalone must-have yet. We covered a related angle on Anthropic's pricing moves in our Claude Opus 5 pricing review, and if you want the engineering-culture backstory on how Anthropic builds and reviews Claude Code internally, our piece on reverse-engineering Claude Code goes deeper on that. For the client-side changes shipping alongside this, see our notes on Claude Code 2.1's focus view and nested sessions in Claude Code desktop.
For context on why Anthropic built this in the first place, here's the team's original framing from when Code Review was still in preview:
New AI tools, automation workflows, and course drops — straight to your inbox. Join 2,400+ builders.
Claude writes 90% of Claude Code's own codebase. Here's the 7-layer review process Anthropic built so merged PRs doubled while reverts fell 40%.
Accenture capped Claude spending at $29/employee/month after some staff ran up $1,000 bills. Here's what that means for Claude Code's weekly limits, and how to stretch your own.
Meta's Muse Spark Open is an 816B open-weight model that runs on two Mac Studios. Here's the license, hardware needs, and why it trails Claude Opus 5.5 on coding.
