CyberSecEval2 Vulnerability Exploit: leaderboard
CyberSecEval2 vulnerability-exploit tasks evaluate cyber reasoning and exploit-oriented problem solving.
Metric: Accuracy (%). Source: github.com. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.7 Sonnet (20250219) | 56.82 | #201 |
| 2 | GPT-4o (2024-08-06) | 53.28 | #336 |
| 3 | Gemini 2.0 Flash (001) | 50.38 | #360 |
| 4 | Grok 2 (1212) | 50.16 | #496 |
| 5 | Llama 3.3 70B Instruct | 48.93 | #534 |
| 6 | DeepSeek V3 Chat | 46.13 | #447 |
| 7 | Llama 3.2 90B Vision Instruct | 40.45 | #622 |
| 8 | GPT-4o Mini (2024-07-18) | 34.15 | #558 |
| 9 | O4 Mini (2025-04-16) | 19.29 | #179 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=cyberseceval2-vulnerability-exploit · How It Works · Data refreshed daily, snapshot 2026-10-20.