← Benchmarks
CyberGym
CyberGym is a benchmark for evaluating AI agents on cybersecurity tasks, testing their ability to identify vulnerabilities, perform security analysis, and complete security-related challenges in a controlled environment.
id cybergym · max 1 · 20 models reported
| # | Model | Score |
|---|---|---|
| 1 | MiMo-V2.6-Flash Xiaomi · open | 0.95 |
| 2 | MiMo-V2.6-Pro Xiaomi · open | 0.94 |
| 3 | DeepSeek-V4.1-Flash DeepSeek · open | 0.88 |
| 4 | Atria Dawn Preview Shanghai AI Laboratory · open | 0.86 |
| 5 | Gemini 3.8 Flash Cyber Google | 0.86 |
| 6 | GLM-5.3 Zhipu AI · open | 0.84 |
| 7 | DeepSeek-V4-Pro-0813 DeepSeek · open | 0.83 |
| 8 | Gemini 3.5 Flash Cyber Google | 0.83 |
| 9 | Claude Mythos Preview Anthropic | 0.83 |
| 10 | GPT-5.5 OpenAI | 0.82 |
| 11 | Grok 4.7 xAI | 0.80 |
| 12 | Claude Opus 4.8 Anthropic | 0.79 |
| 13 | Hy4 preview Tencent · open | 0.78 |
| 14 | DeepSeek-V4-Flash-0731 DeepSeek · open | 0.77 |
| 15 | Claude Opus 4.6 Anthropic | 0.74 |
| 16 | Claude Opus 4.7 Anthropic | 0.73 |
| 17 | Seed 2.1 Pro ByteDance | 0.70 |