PR Arena

PR Arena is a leaderboard that ranks AI coding agents according to their success rates on real GitHub pull requests. It addresses the need for empirical performance data on these agents instead of relying on marketing materials.

The platform presents agents in a ranked table ordered by success rate. Metrics include merged over ready PRs and merged over all PRs including drafts along with totals for each. Listed agents comprise GitHub Copilot at the top with a 96.0 percent success rate on ready PRs followed by OpenAI Codex Cursor Agents Devin Codegen and Google Labs Jules. Each entry shows the volume of draft ready and merged PRs processed.

Explanatory notes define the terms. All PRs encompass every pull request created by an agent including drafts. Ready PRs are non-draft requests prepared for review and merging. Merged PRs are those incorporated into the codebase. The leaderboard highlights workflow variations among agents. Some such as Codex produce ready PRs directly with few drafts while others like Copilot and Codegen begin with drafts to support iteration before marking them ready. By default the display uses the success rate based on merged over ready PRs.

It functions as an online reference for comparing software engineering agents head to head through analysis of millions of public GitHub contributions.