Do AI agents ship code that actually ships?
Live PR-merge leaderboard for the eight major AI coding agents — Claude Code, GitHub Copilot, Cursor, OpenAI Codex, Devin, Aider, Gemini, OpenAI. 30-day rolling window of real public GitHub data. Updated every 30 minutes.
No PRs detected yet — first fetch in progress. Reload in ~2 minutes.
No GitHub token detected — the GitHub Search API is rate-limited to 10 req/min, so the dataset may be sparse until a token is injected.
Leaderboard — 30-day window
| # | Agent | PRs | Merged | Open | Closed (no merge) | Merge rate | Median TTM | p90 TTM | Stale rate | Change-req |
|---|---|---|---|---|---|---|---|---|---|---|
| Loading… | ||||||||||
Merge rate = merged / (merged + closed unmerged). Open PRs are excluded from the denominator so a burst of new opens does not tank the score. TTM = time from PR open to merge, in minutes. Stale rate = open PRs older than 14 days / open PRs. Change-req = PRs with at least one CHANGES_REQUESTED review / total PRs.
Merge rate per agent
Median time-to-merge (lower = better)
Agent
Time-to-merge histogram
Recent PRs
| Title | Repo | State | Created | Merged at | +/- |
|---|