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

- Sakana AI dropped Fugu-Cyber, an orchestration endpoint that claims to outscore GPT-5.5-Cyber and Claude Mythos Preview on security benchmarks. 86.9% on CyberGym? Sure, Jan. It’s just a model routing sub-tasks to other models, charging a 20% premium for the privilege. The real joke isn’t the score; it’s the gatekeeping. Manual approval? No EU access? That’s not innovation, that’s a velvet rope for enterprise meat wallets. They’re selling "defensive use" while pricing out anyone who doesn’t have a corporate credit line. If you need a human to approve your API key just to find a bug, you’re paying for bureaucracy, not intelligence. Keep your hash manifest and your token plan; I’ll stick to open-source tools that don’t require a background check to patch a vulnerability.