Anthropic launches AI-powered Code Review tool for Claude Code at $15-25 per review

Mar 9, 2026

Anthropic has launched Code Review, an AI-powered tool for its Claude Code platform that uses multiple AI agents to find bugs in software pull requests. The tool addresses the review bottleneck created by AI-generated code. It costs $15-25 per review, a price some developers criticize as high. Internal tests show it triples substantive feedback and has a sub-1% error rate. The tool is available for Team and Enterprise customers.

Claude Code Review launch

▪The launch coincides with Anthropic filing two lawsuits against the U.S. Department of Defense over a supply chain risk designation.
▪The tool is a more thorough and expensive option than the existing open-source Claude Code GitHub Action.
▪Anthropic launched its AI-powered "Code Review" tool for Claude Code on March 9, 2026.
▪Code Review is available in a research preview for Claude Code Team and Enterprise subscribers.
▪The tool is designed to spot bugs and logic errors in AI-generated code before a human reviewer sees it.

Pull request workflow context

▪A pull request is a workflow where developers submit code changes for review before they are merged into a main software codebase.
▪The tool was created to address the review bottleneck caused by an increase in pull requests from developers using AI coding assistants.
▪The rise of "vibe coding," using AI to generate code from plain language, has accelerated development but also introduced new bugs.

Multi-agent review system architecture

▪The system is designed to focus on identifying logical errors rather than stylistic issues to provide more actionable feedback.
▪Identified issues are labeled by severity with colors: red for most severe, yellow for potential problems, and purple for pre-existing issues.
▪An average review takes about 20 minutes to complete, with the time scaling based on the pull request's complexity.
▪Code Review uses multiple AI agents working in parallel, with each agent examining the code from a different perspective.
▪A final agent aggregates and ranks the findings, removes duplicates, and prioritizes them before presenting them as a single comment.

Bug detection test results

▪During internal use, Anthropic engineers marked less than 1% of the tool's findings as incorrect.
▪For large pull requests with over 1,000 changed lines, the tool found issues 84% of the time, with an average of 7.5 issues per review.
▪In one test, the tool flagged a single-line change that would have broken a production service's authentication mechanism.
▪In internal testing at Anthropic, the use of Code Review increased the percentage of pull requests receiving substantive comments from 16% to 54%.

Per-review pricing structure

▪The cost of a review scales with the size and complexity of the pull request being analyzed.
▪Code Review is billed on token usage and costs an average of $15 to $25 per review.