Sakana AI releases multi-agent system Fugu: surpassing GPT 5.4 and Opus 4.6 in performance

Golden Finance reports that Japan's AI development company Sakana AI announced the launch of its flagship commercial product Sakana Fugu and opened early beta testing applications.
The system is positioned as a multi-agent orchestration system, providing services in the form of a single-model API compatible with OpenAI formats, including Sakana Fugu Mini optimized for low latency and Sakana Fugu Ultra designed for demanding tasks.
The product architecture is based on the Trinity and Conductor papers published by the team at ICLR 2026.
The core of the system is a self-learning lightweight language model that does not rely on manually preset team roles or fixed pipelines, but dynamically calls Worker model pools and assigns tasks based on the difficulty of the task.
The system also supports scaling during testing and has adaptive recursive calling capabilities.
The model can read its own previous outputs as context, enabling it to autonomously identify flaws during operation and initiate correction workflows, with users able to set the recursive depth as a tunable computational axis during inference.
Evaluation data shows that Sakana Fugu Ultra outperforms cutting-edge single models in hardcore reasoning and coding benchmarks.
In GPQAD, LCBv6, and SWEPro tests, Sakana Fugu Ultra achieved scores of 95.1, 93.2, and 54.2 respectively, outperforming mainstream flagship single models such as GPT 5.4, Gemini 3.1, and Opus 4.6.
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pinned