VEX-Bench - Status Accuracy: leaderboard
LLM agents deciding whether a software-supply-chain vulnerability is actually exploitable in a repository (VEX status), in Docker-backed environments under a coding-agent harness; binary accuracy, mean of three runs. Higher is better. Source: arxiv.org/abs/2609.08040.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 89.3 | #33 |
| 2 | Claude Opus 4.6 | 88.4 | #67 |
| 3 | GLM-5.1 | 84.4 | #109 |
| 4 | DeepSeek V4 Pro | 84 | #112 |
| 5 | GPT-5.4 Mini | 82.2 | #228 |
| 6 | Claude Sonnet 4.6 | 81.8 | #94 |
| 7 | Kimi K2.6 | 81.8 | #113 |
| 8 | DeepSeek V4 Flash | 72.9 | #155 |
| 9 | MiniMax-M2.7 | 71.1 | #270 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=vex-bench-status-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-05.