Sakana Fugu: Matches Fable & Mythos on AI Benchmarks

Sakana AI has launched Sakana Fugu, an AI system that coordinates multiple existing models to achieve high performance. This Tokyo-based lab, valued at over two-and-a-half billion dollars, aims to match top-tier AI capabilities without building a single powerful model. Here's the thing: Fugu dynamically assembles agents from a pool and routes work between them, learning efficient patterns. This system delivers results through a single OpenAI-compatible API. What's interesting is its benchmark performance. On SWE-Bench Pro, Fugu Ultra scored 73.7, surpassing Claude Opus 4.8 and GPT-5.5. It also led on LiveCodeBench, and on Humanity’s Last Exam, Fugu Ultra essentially matched Opus 4.8. Sakana states Fugu performs comparably to high-tier models like Claude Fable 5 and Mythos Preview on several benchmarks. The bottom line is that this innovative approach could offer a new way to achieve advanced AI performance by orchestrating existing models.