SWE-PRBench — leaderboard
Pull-request review benchmark with a public paper-baseline leaderboard for model review quality and false-positive behavior.
Metric: Overall (sbar) (self-reported). Source: benchmarklist.com. Status: saturation imminent. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Haiku 4.5 | 15.3 |
| 2 | Claude Sonnet 4.6 | 15.2 |
| 3 | DeepSeek V3 | 15 |
| 4 | Mistral Large 3 | 14.7 |
| 5 | GPT-4o | 11.3 |
| 6 | GPT-4o Mini | 10.8 |
| 7 | Mistral Small | 10.6 |
| 8 | Llama 3.3 70B Instruct | 7.9 |
Interactive version: theaggregate.ai/benchmark?slug=swe-prbench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.