HO1 runs tasks that other models refuse.
Hinoki Offensive One works through the real task turn by turn within the policy you set. GPT-6.1-Sol refuses outright.
Transcript
operator@range:~$
- HO1 Hinoki
- Completed in 6 turns
operator@range:~$
- GPT-6.1-Sol OpenAI
- Refused the task
Scores like a gated cyber model. Open to anyone.
On CyberGym, HO1 scores 88.4%, 2.8 points above OpenAI’s dedicated GPT-5.5-Cyber.
- HO1Hinoki, open to anyone0.0%
- GPT-5.5-CyberOpenAI’s latest Daybreak model, gated0.0%
- GPT-5.6-SolOpenAI’s latest general model0.0%
- DeepSeek-V4-ProDeepSeek’s top open-weight model0.0%
- Claude Mythos PreviewAnthropic, preview access0.0%