Fugu is not a single model but Sakana AI's multi-agent orchestration system, trained to route each request to the most appropriate specialized model. It distributes Thinker, Worker, and Verifier roles across several models and coordinates their results. The model pool is not fixed — it is selected dynamically per task, and customers can exclude specific providers or models. Where Fugu Ultra prioritizes accuracy on hard problems, Fugu is the everyday default that balances quality against latency. It has no fixed token price. Billing follows the rate of the highest-tier model active at the time, and fees do not stack when multiple agents run together.
API|비전심층 사고웹 검색|상용 모델
전문가 성능 평가
코딩
⚠
LiveCodeBench
93%↑23%
Sakana Namazu