Fugu is Sakana AI's orchestration model rather than a single base language model: a single endpoint dynamically routes incoming requests to a pool of specialized agents so the system behaves like one model while drawing on multiple experts under the hood. Users can opt specific agents out of that pool to satisfy data, privacy, and compliance constraints. Sakana positions Fugu as the everyday default, balancing strong performance with low latency, while a sibling variant called Fugu Ultra coordinates a deeper pool of expert agents and routes between one to three of them depending on the problem, trading higher cost for stronger answers on hard, high-stakes tasks.

The Fugu family also includes Fugu-Cyber, a specialized orchestration endpoint released in July 2026 that targets cybersecurity workflows such as security analysis, vulnerability research, and threat investigation. On industry cyber benchmarks Sakana reports a success rate of 86.9% on CyberGym, which tests analyzing complex codebases to verify real-world vulnerabilities, and 72.1% on CTI-REALM, which measures turning raw threat intelligence reports into working detection rules, placing it in the same range as cyber-focused frontier models. Fugu-Cyber is served only through pay-as-you-go billing and requires an access request form, reflecting the sensitivity of its target use cases.