Sakana AI has launched Fugu-Cyber, an orchestration model specifically tuned for cybersecurity tasks. This model achieved 86.9% on the CyberGym benchmark and 72.1% on the CTI-REALM benchmark, positioning it as a competitive option against other specialized models like GPT-5.5-Cyber and Claude Mythos Preview. Fugu-Cyber operates within Sakana's Fugu orchestrator, which delegates tasks to specialized models, with a particular emphasis on security-focused validation steps. AI
IMPACT This specialized cybersecurity model sets new benchmarks, potentially influencing the development of AI agents for security tasks.
RANK_REASON Sakana AI is a frontier lab releasing a new model with benchmark results. [lever_c_demoted from frontier_release: ic=2 ai=1.0]
- Claude Mythos Preview
- Conductor
- CTI-REALM
- Fugu
- Fugu-Cyber
- GPT-5.5-Cyber
- ICLR 2026
- KQL
- Microsoft
- MITRE ATT&CK
- Sakana AI
- Sigma
- TRINITY
- UC Berkeley
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →