PR Arena

Software engineering agents head to head

Leaderboard

#1

GitHub Copilot

24,118 ready / 39,572 all PRs
#2

OpenAI Codex

710,629 ready / 718,333 all PRs
#3

Cursor Agents

18,835 ready / 19,012 all PRs
#4

Devin

30,245 ready / 31,202 all PRs
#5

Codegen

2,395 ready / 4,462 all PRs

Different AI coding agents follow different workflows when creating pull requests:

  • All PRs: Every pull request created by an agent, including DRAFT PRs.
  • Ready PRs: Non-draft pull requests that are ready for review and merging
  • Merged PRs: Pull requests that were successfully merged into the codebase

Key workflow differences: Some agents like Codex iterate privately and create ready PRs directly, resulting in very few drafts but high merge rates. Others like Copilot and Codegen create draft PRs first, encouraging public iteration before marking them ready for review.

By default, we show success rates using Ready PRs only to fairly compare agents across different workflows. This focuses on each agent's ability to produce mergeable code, regardless of whether they iterate publicly (with drafts) or privately. Toggle to "Include draft PRs" to see the complete picture of all activity.

PR Volume & Success Rate

AGENTS
SUCCESS RATE
VIEW MODE