Sakana AI launches Fugu multiagent system rivaling top models

A Tokyo-based AI startup, Sakana AI, has launched a new multiagent orchestration system called Fugu, designed to deliver performance rivaling top-tier frontier models while offering users a hedge against vendor lock-in and geopolitical export controls. The system, which became available on June -22, 2026, operates through a single OpenAI-compatible API.

Fugu functions as a conductor model, dynamically routing user requests to a swappable pool of specialized AI agents. The system handles selection, delegation, verification, and synthesis internally, presenting the user with a single coordinated output. Sakana is offering two variants: a base Fugu model optimized for low latency and everyday tasks, and Fugu Ultra, which uses a deeper pool of agents for complex, multi-step problems like AI research, cybersecurity analysis, and scientific paper reproduction.

The launch follows the suspension of public access to Anthropic's powerful Fable 5 and Mythos 5 models on June 12, 2026, due to a U.S. export control directive. Sakana CEO David Ha, a former Google Brain researcher, positioned Fugu as a direct response to such risks. He argued that relying on a single provider's AI for critical...