Tokyo-based AI startup Sakana AI has announced the release of Fugu Max and Fugu Ultra v2, two updates to its Fugu orchestration system. The Fugu series is designed to orchestrate multiple closed and open-weights models to perform specific tasks, aiming to outperform single models by utilizing the most efficient model for each requirement.

Fugu Ultra v2 is positioned as a high-performance version that pushes the boundaries of autonomous execution. According to Sakana AI, it achieves superior performance in complex tasks such as coding and visual reasoning. For example, it achieved a score of 48.3 on the Chartography benchmark, outperforming Claude Fable 5 and Opus 5. Notably, Fugu Ultra v2 achieves these results without relying on models such as Claude Fable 5.1 or GPT-6 Astra in its agent pool.

Fugu Max focuses on expanding cost efficiency and the range of available models. By integrating a wider variety of open-weights and specialized models, including the NVIDIA Nemotron family, it allows users to achieve frontier-grade results at a significantly lower cost than single-model providers.

Both Fugu Max and Fugu Ultra v2 are available via an OpenAI-compatible API. For existing users of the Fugu architecture, upgrading requires only a single-line parameter change.

Sources: