Sakana AI Releases Fugu-Cyber: An Orchestration Model Reporting 86.9% on CyberGym and 72.1% on CTI-REALM

Sakana AI released Fugu-Cyber, a cybersecurity-focused orchestration model that coordinates multiple specialist AI models to solve security tasks. The model reported performance scores of 86.9% on CyberGym, a benchmark testing vulnerability discovery and proof-of-concept generation, and 72.1% on CTI-REALM, a benchmark testing threat detection rule creation from security reports. These scores represent marginal improvements over comparable frontier models, with access restricted through manual application review, acceptable use policies limiting defensive use, and pricing at a 20% premium over the base Fugu model. The orchestration approach uses specialized roles for different stages of security analysis, with verification emphasized as the critical component before any patches are proposed.