Anthropic launches Code Review service at $15-25 per PR, takes 20 minutes per review

Anthropic has introduced Code Review, a hosted service for teams and enterprise customers that automates source code analysis using multiple AI agents working in concert to identify bugs and vulnerabilities in GitHub pull requests.

Service Pricing and Performance

Code Review operates on a per-pull-request billing model. Anthropic reports that reviews average $15-25 per PR, with costs scaling based on pull request size and complexity. The service requires approximately 20 minutes per review on average, though this varies with code scope.

For context, Code Rabbit—a competing AI code review service—charges $24 monthly. At Anthropic's pricing, a team conducting three code reviews daily would spend roughly $1,350 monthly, compared to a flat $24 subscription for alternatives.

Detection Rates and Issue Identification

According to Anthropic's internal testing, detection accuracy varies significantly by pull request size:

• Large PRs (>1,000 lines changed): 84% receive comments, averaging 7.5 issues flagged
• Small PRs (<50 lines changed): 31% receive comments, averaging 0.5 issues flagged

Anthropinc claims human developers reject fewer than one percent of issues identified by Code Review.

How It Works

Code Review analyzes GitHub pull requests and posts findings as inline comments. According to Anthropic, a fleet of specialized agents examine code changes within the full codebase context, searching for logic errors, security vulnerabilities, broken edge cases, and subtle regressions.

The service integrates with existing CI/CD pipelines, building on Anthropic's existing Claude Code GitHub Action that performs lighter-weight reviews on demand.

Real-World Results

Anthropic reports that TrueNAS's open-source middleware project discovered a critical bug during ZFS encryption refactoring—a type mismatch that risked erasing the encryption key cache during sync operations. In another internal case, Code Review caught a single-line change to a production service that would have broken authentication without being caught by human review.

Analysis and Limitations

The service's 20-minute turnaround and $15-25 per-PR cost raise practical questions for development teams. At typical developer hourly rates ($60-150+), outsourcing code review to an AI service that takes 20 minutes must provide value beyond what human review delivers in the same timeframe—particularly for smaller pull requests that already average under one issue.

The token-intensive nature of running multiple agents explains both the cost and latency. This represents a trade-off between thoroughness (multiple agent perspectives) and practicality (time and expense).

What This Means

Anthropicis monetizing code review as a premium service for organizations that can absorb per-PR costs in exchange for automated second-pair-of-eyes coverage. The approach targets enterprise teams where catching critical bugs justifies the expense. However, the 20-minute delay and $15-25 fee-per-review make it most viable for asynchronous review workflows rather than pre-merge gating. Success will depend on whether the bug-detection rate justifies the cost compared to traditional developer code review or cheaper alternatives.