Benchmark

CyberGym

Offensive-security capture-the-flag style tasks.

Score history →

Model scores

#ModelScore
1DeepSeek V4.1-Flash 091088.1%
2GPT 5.6-Sol84.5%
3GLM 5.384.5%
4DeepSeek V4-Pro 081383.3%
5MiniMax M3.1 Flash82%
6Kimi K380%
7DeepSeek V4-Flash 073176.7%