CyberGym is a large-scale cybersecurity evaluation framework for assessing AI agents on real-world vulnerability analysis tasks in reproducible execution environments.
Browse the latest scores, model modes, release dates, and parameter sizes for CyberGym.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
| Rank | Model | License | |||
|---|---|---|---|---|---|
— |  DeepSeek-V4.1-Flash Thinking Level · MaxTools | 88.10 | 2026-09-10 | 552B | Free Commercial |
— |  GPT-5.6 Sol Thinking Level · MaxTools | 84.50 | 2026-06-26 | Unknown | Closed |
— |  GLM-5.3 Thinking Level · MaxTools | 84.50 | 2026-08-14 | 744B | Conditional |
— | DeepSeek-V4-Pro Thinking Level · Extra HighTools | 83.30 | 2026-08-13 | 1600B | Free Commercial |
— |  Kimi K3 Thinking Level · MaxTools | 80.00 | 2026-07-16 | 2800B | Conditional |
— |  Hy4 preview Thinking Level · HighTools | 78.40 | 2026-08-28 | 770B | Free Commercial |
— | DeepSeek-V4-Flash Thinking Level · MaxTools | 76.70 | 2026-04-24 | 284B | Free Commercial |
— | DeepSeek-V4-Flash-Vision-Exp Thinking Level · MaxTools | 75.30 | 2026-08-21 | 305B | Free Commercial |
