Sakana Fugu: New AI Rivals Frontier Models

Sakana AI Labs has unveiled Sakana Fugu, a new multi-agent orchestration system. This system coordinates specialist AI models through a single API. What's interesting is that it claims benchmark scores that rival top models from other companies. For example, Fugu Ultra, the system’s top-tier variant, scored 73.7 on the SWE-Bench Pro benchmark. This performance matches or approaches offerings like Anthropic’s Fable 5 and Mythos Preview. Fugu works as a trained conductor model, managing an orchestra of specialist AI agents. It dynamically handles routing, role assignments, and result synthesis, learning collaboration patterns autonomously. The system is delivered through a single API endpoint, compatible with OpenAI's format, making integration straightforward for developers. This approach also helps address concerns like export control risk and vendor lock-in by orchestrating models from various providers. The bottom line is this system could change how businesses integrate and manage AI, offering flexibility and powerful performance.