A new data directory launched this week attempts to cut through the marketing noise in the AI code review space by ranking 56 tools based on hard, reproducible metrics. The index, hosted on Cloudflare Pages, tracks public GitHub pull request activity for week 2026-W39 (September 21โ€“27) and links every number to the specific query used to generate it. This transparency is a welcome shift in a sector often dominated by vague 'users served' claims, providing developers with a concrete baseline for evaluating which bots are actually active in public repositories.

The Heavy Hitters: Copilot and Codex Lead

Unsurprisingly, the giants of the AI infrastructure world are pulling the most public volume. GitHub Copilot code review tops the list with 146,488 public PRs reviewed, followed closely by OpenAI Codex code review with 115,760. These numbers reflect the massive default adoption within the GitHub and Azure DevOps ecosystems. However, the data reveals a critical nuance: while Copilot and Codex have high review counts, their 'PRs commented' metrics are nearly identical to their review counts, suggesting they are primarily engaging with automated or low-effort comments rather than deep, multi-turn discussions.

The Rise of Specialized and Self-Hosted Tools

Beyond the hyperscalers, specialized tools like CodeRabbit (81,094 reviews) and Greptile (30,195) are carving out significant niches. CodeRabbit stands out for its multi-platform support, covering GitHub, GitLab, Bitbucket, and Azure DevOps, though it requires an enterprise plan for self-hosting. For builders prioritizing data sovereignty, LlamaPReview is a notable entry at rank 13 with 1,095 reviews, offering a self-hosted option with a free tier. The index also highlights tools like cubic and Pullfrog, which show healthy engagement relative to their market presence, indicating that smaller, focused tools can still generate meaningful public activity.

Limitations of Public Metrics

The index is explicit about its blind spots. A 'zero' in the public PR column does not mean a tool is unused; it may indicate the tool posts under a shared identity like github-actions[bot] or a user's own account, making attribution impossible. Tools like Claude Code Review and Cursor Bugbot fall into this 'shared identity' bucket, meaning their actual impact is likely much higher than the visible data suggests. Furthermore, the metric ignores private repositories, unique users, and code quality improvements, meaning a tool with zero public PRs could still be the best choice for a closed-source enterprise.

Key Takeaways

  • GitHub Copilot and OpenAI Codex dominate public volume with over 100,000 PRs reviewed each in a single week.
  • CodeRabbit and Greptile lead the independent vendor space, with CodeRabbit offering broader platform support.
  • Self-hosted options like LlamaPReview and Bito AI are gaining traction among privacy-conscious developers.
  • Public PR counts are an imperfect proxy for market share due to shared bot identities and private usage.

The Bottom Line

This index is a useful starting point for shortlisting tools, but don't mistake public volume for product quality. Use the reproducible queries to verify claims, but test these bots in your own private repos before trusting the leaderboard.