Fugu Max

Fugu Max was released on 11 September 2026 alongside Fugu Ultra v2, and Sakana AI is explicit that the thing being shipped is not a model: "Fugu is an architecture, not a single model." Both releases share what the company calls "the same core orchestration architecture optimized for two distinct missions" — Fugu Max answers "what is the best possible output we can deliver at the lowest possible cost?", while Ultra v2 goes for maximum capability. The architecture "dynamically routes tasks to the leanest model capable of solving them", drawing on open-weights and specialist models including NVIDIA Nemotron models through a partnership. Pricing is $2 per million input tokens and $6 per million output. It is available through an OpenAI-compatible API, and Sakana says existing Fugu users switch with "a single-line parameter change". Sakana names Terminal Bench 2.1, GPQAD, AA-LCR, GDP.pdf, AutomationBench and SWEFish as benchmarks Fugu Max is best overall on, but publishes no figure for any of them in that announcement — so the claim is recorded here in prose and no score is stored against it.