Cybergym

Cybergym - agentic cybersecurity benchmark evaluating autonomous hacking and security task capabilities.

Category

General Agents

Max Score

100.0

Score Type

percent

Active

Yes

Model Scores

Scores for Cybergym

Model Score Date Verified
DeepSeek-V4-Flash-0731
76.92%
31.07.2026 Verified
Opus-4.8
89.88%
31.07.2026 Verified
DeepSeek-V4-Flash
38.70
13.08.2026 Verified
DeepSeek-V4-Pro
28.34%
13.08.2026 Verified
DeepSeek-V4-Pro-0813
90.28%
13.08.2026 Verified
Kimi K3
83.60%
13.08.2026 Verified
Fable-5
89.88%
13.08.2026 Verified
GLM-5.3
92.71%
28.08.2026 Verified
GLM-5.2
77.94%
28.08.2026 Verified
Qwen3.8-2.4T-A95B
80.57%
28.08.2026 Verified
GPT-5.6 Sol
90.89%
28.08.2026 Verified
Hy3
26.52%
28.08.2026 Verified
Hy4 preview
80.36%
28.08.2026 Verified
DeepSeek V4 Pro
90.28%
28.08.2026 Verified
DeepSeek-V4-Flash-Vision-Exp
74.09%
08.09.2026 Verified
DeepSeek-V4.1-Flash
100.00%
10.09.2026 Verified