Sakana AI’s Fugu is more an AI coordinator than a new model
Sakana AI released Fugu and Fugu Ultra on June 22. Fugu is not a new . It is an orchestrator that uses a to choose from a pool of other public , call them, check their work, and combine the results behind one .
The better way to understand it is not as Sakana’s direct GPT rival, but as a learned router and coordinator sold as a single API. Sakana says the approach is grounded in two ICLR 2026 papers, TRINITY and Conductor. In Sakana’s own benchmark numbers, Fugu Ultra scored 73.7 on SWE-Bench Pro, ahead of Opus 4.8 at 69.2, GPT-5.5 at 58.6, and at 54.2.
Sakana also reported 95.5 on GPQA-D, 93.2 on LiveCodeBench, and 82.1 on . These are vendor-reported numbers, not independently checked, and Fugu Ultra still trails Fable 5, which it cannot include in its model pool.
Key points
- Fugu is an orchestrator, not a new .
- It works through one while using several public behind the scenes.
- It handles model choice, delegation, checking, and final synthesis internally.
- Sakana’s own numbers show Fugu Ultra ahead of several named models on coding and reasoning benchmarks.
- The reported scores are not independently verified, and Fugu Ultra still trails Fable 5.