Informal comparison
IronBee raw-model code review
Compare raw language models reviewing the same real pull requests.
Published leaderboard
Explore scores and evaluation settings on the publisher’s leaderboard.
Open leaderboardCoverage
21 models, 58 configurations, three repeats; a 50-PR main set and a separate 10-PR frontier subset.
Methodology
Rerun Martian Code Review Bench with a common prompt and a calibrated three-model judging panel. Report precision, recall, F2, cost and latency.