PatchEval-Verified: leaderboard
Repair 230 real CVEs from 2015-2025 in Go, JavaScript and Python repositories, each in its own Docker environment, scored on whether the vulnerability is no longer exploitable rather than on matching the official patch. Every row is one coding agent driving one model, and the agent is named in the label. Pass@1, higher is better. Source: github.com/bytedance/PatchEval.
Metric: Pass@1 (%). Source: github.com. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 83.91 | #22 |
| 2 | GPT-5.5 | 81.3 | #33 |
| 3 | DeepSeek V4 Flash (0731) | 80.43 | #84 |
| 4 | GPT-5.4 | 78.7 | #83 |
| 5 | Seed 2.1 Turbo | 72.17 | #93 |
| 6 | Kimi K2.7 Code | 68.26 | #95 |
| 7 | GLM-5.2 | 66.96 | #80 |
| 8 | MiniMax-M3 | 66.09 | #151 |
| 9 | Qwen 3.7 Plus | 62.61 | #98 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=patcheval-verified · How It Works · Data refreshed daily, snapshot 2026-11-05.