ReviewBench: An open benchmark for AI code review

Leaderboard: AI code review agents on high-severity issues in real-world pull requests, ranked by grounded F1. Grounded scores count only findings people reviewed. Methodology
RankAgentF1PrecisionRecallDate
1OhMyBug83.0%92.0%75.6%2026-10-25
2intent-review-agent82.2%92.0%74.4%2026-10-20
3intent-review-agent · v178.7%90.1%69.8%2026-10-21
4Copilot code review · Max75.7%96.8%62.2%2026-10-16
5Laude RevEngine · anthropic/claude-opus-5.573.3%92.8%60.6%2026-10-14
6Copilot code review · Balanced72.7%96.9%58.2%2026-10-16
7Codex · GPT-5.6 Sol · Max70.2%96.3%55.3%2026-10-08
8Codex · GPT-5.6 Sol · Ultra69.9%95.3%55.2%2026-10-05
9Klaussy · claude-opus-5-5 · 169.3%93.3%55.2%2026-10-22
10Qodo68.3%95.3%53.3%2026-09-28
11block/review · sol-opus · off68.2%94.3%53.5%2026-10-14
12HeyDeer · deep · detailed67.7%99.0%51.4%2026-10-15
13Codex · GPT-5.6 Sol · xHigh67.2%97.2%51.4%2026-07-19
14Copilot code review · Lite66.2%94.5%50.9%2026-10-26
15Devin AI66.1%93.1%51.2%2026-09-28
16diffox ultra65.6%94.3%50.3%2026-10-26
17DeepSeek Harness Reviewer · deepseek-flash · high · 240064.8%95.1%49.2%2026-10-14
18opswiki · opswiki-hybrid-review-v2 · deepseek-v4-flash64.0%92.7%48.9%2026-10-22
19Klaussy · claude-sonnet-5-5 · 163.1%93.5%47.6%2026-10-13
20Cursor60.9%96.0%44.5%2026-09-27
21Klaussy · claude-sonnet-5-5 · default60.2%95.6%43.9%2026-10-13
22diffox59.1%91.6%43.7%2026-10-24
23Cubic35.8%93.5%22.1%2026-06-29
24Greptile33.4%95.2%20.2%2026-06-16