SWE-PRBench (Config C): leaderboard
Metric: Composite review score (0-1, scaled to 0-100): per pull request, 0.40 recall + 0.25 precision + 0.15 alignment + 0.10 actionability + 0.05 efficiency minus hallucination, redundancy and excess-plausible penalties, clamped to 0-1 and averaged with weights log(human comments + 1), on a stratified 100-PR sample of SWE-PRBench (40 direct, 40 contextual and 20 latent-difficulty pull requests from active open-source repositories), each model asked at temperature 0 for 4 to 6 line-anchored review comments as JSON, a GPT-5.2 judge matching them to the human reviewers' comments (confirmed, plausible or fabricated), with the diff, execution context, behaviour mapping and test signatures (about 2,500 tokens); higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Mistral Large 3 | 13.5 | #388 |
| 2 | Claude Haiku 4.5 | 12.6 | #271 |
| 3 | Claude Sonnet 4.6 | 12.2 | #85 |
| 4 | DeepSeek V3 | 11.8 | #312 |
| 5 | GPT-4o Mini | 9.3 | #588 |
| 6 | Mistral Small | 9.1 | #700 |
| 7 | GPT-4o | 9 | #333 |
| 8 | Llama 3.3 70B | 6.5 | #569 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=swe-prbench-config-c · How It Works · Data refreshed daily, snapshot 2026-10-11.