Zhun Wang et al.

CyberGym

A cybersecurity task benchmark for evaluating defensive cyber workflows and vulnerability-oriented agent performance.

In index
Agentic
Best in agentic
Source ↗

Models29

Top result95.1%MiMo-V2.6-Flash

Top-3 spread7.0 pts

Year2026

Result and price

25 of 29 results

Results

29 results

#ModelResult

01MiMo-V2.6-Flash95.1%

02MiMo-V2.6-Pro94.0%

03DeepSeek V4.1 Flash 101788.1%

04Ling 3.1 Flash87.9%

05Fugu Cyber86.9%

06Atria Dawn Preview86.5%

07Gemini 3.8 Flash Cyber86.2%

08Step 5 Preview84.7%

09GPT-5.6 Sol84.5%

09GLM-5.384.5%

11Claude Mythos 583.8%

12DeepSeek V4 Pro 081383.3%

13Gemini 3.5 Flash Cyber83.2%

14Claude Mythos Preview83.1%

15MiniMax M3.1 Flash82.2%

16GPT-5.581.8%

16GPT-5.6 Terra81.8%

18GPT-5.479.0%

19Hy4 preview78.4%

20GPT-5.6 Luna77.9%

21DeepSeek V4 Flash 073176.7%

22Claude Opus 4.7 (Adaptive)73.1%

23GLM-5.168.7%

24Claude Opus 4.666.6%

25Claude Sonnet 4.665.2%

26Muse Spark 1.159.0%

27Claude Opus 4.550.6%

28Muse Spark43.5%

29GLM-543.2%