Anthropic launches a code review tool with multi-agent AI bug detection, but it's pricey

Anthropic has launched Code Review, an AI-powered system that deploys a team of agents to inspect every pull request for bugs typically missed by automated scans. Unlike lightweight solutions, Code Review emphasizes depth, providing a more rigorous option than the open source Claude Code GitHub Action, which remains available.

When a pull request is opened, the system assigns several agents who independently examine the code in parallel. These agents collaborate to detect potential issues, verify findings to minimize false positives, and prioritize bugs by severity. The review concludes with a concise, high-signal summary on the pull request, complemented by detailed inline comments for specific issues.

Following this process, Code Review adapts the review depth to match the complexity of each pull request: larger or more involved changes receive a thorough inspection with additional agents, while simple edits get a lighter assessment. Anthropic's internal data indicates that the average review takes about 20 minutes to complete. Reviews are billed on token usage and typically cost between $15 and $25, with expenses scaling by the size and complexity of the code change.

As of today, Code Review is available in beta as a research preview for customers on Anthropic's Team and Enterprise plans.