EviRepair Study - EviRepair - PatchEval Python: leaderboard
Vulnerability repair in a matched sandbox with identical tool access. Codex v0.153.2 and OpenCode v1.17.7 get five hours; EviRepair uses three hours diagnosis plus two hours patch synthesis. PatchEval has 230 vulnerabilities and SEC-bench has 300. Scores are percent resolved, with each harness kept separate.
Metric: Resolved (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4.1 Flash | 94.29 | #36 |
| 2 | Qwen 3.8 27B | 78.57 | #96 |
| 3 | GLM-5.2 | 77.14 | #71 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=evirepair-study-evirepair-patcheval-python · How It Works · Data refreshed daily, snapshot 2026-10-16.