Sakana launches Fugu as a multi-agent AI system

Economic Times reported that Tokyo-based Sakana AI has launched Fugu, a multi-agent AI system that presents itself as a single model while dynamically selecting and coordinating specialized models behind the scenes. The article says Fugu is designed to break down prompts, route subtasks across a pool of frontier models, and optimize performance across coding, reasoning, research, and other complex workloads. The system builds on research papers called TRINITY and Conductor, presented at ICLR 2026. Sakana is led by David Ha and Llion Jones, and previously raised a roughly $135 million Series B at a $2.65 billion valuation.

Key details: Published June 23, 2026 at 17:08 IST, Sakana Fugu is described as a multi-agent orchestration system, The system can coordinate specialized models for different subtasks, It builds on TRINITY and Conductor research presented at ICLR 2026.

Why it matters: Fugu points to a different model competition: not just bigger single models, but orchestration layers that choose and coordinate multiple specialized systems.