CyberGym: leaderboard
Cybersecurity agent benchmark for discovering, exploiting, and reasoning about vulnerabilities in controlled challenge environments.
Metric: Score (%). Source: arxiv.org. Source: llm-stats.com. Saturation forecast: Around January 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | MiMo-V2.6-Flash | 95.1 | #70 |
| 2 | MiMo-V2.6-Pro | 94 | #42 |
| 3 | DeepSeek V4.1 Flash | 88.1 | #36 |
| 4 | GLM-5.3 | 84.5 | #39 |
| 5 | DeepSeek V4 Pro (0813) | 83.3 | #53 |
| 6 | Claude Mythos Preview | 83.1 | #4 |
| 7 | MiniMax-M3.1-Flash | 82.2 | #33 |
| 8 | Mistral Large 4 | 82 | #94 |
| 9 | GPT-5.5 | 81.8 | #31 |
| 10 | Grok 4.7 | 80.3 | #25 |
| 11 | Claude Opus 4.8 | 78.8 | #37 |
| 12 | Hy4 preview | 78.4 | #38 |
| 13 | DeepSeek V4 Flash (0731) | 76.7 | #78 |
| 14 | Claude Opus 4.6 | 73.8 | #63 |
| 15 | Claude Opus 4.7 | 73.1 | #49 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=cybergym · How It Works · Data refreshed daily, snapshot 2026-10-20.