Fugu Max, released by Japan's Sakana AI on 11 September 2026, is the cost-optimised branch of the Fugu orchestration engine. Instead of running one large network, it maintains a pool of open-weight and specialised models — including NVIDIA's Nemotron family, added through a partnership announced in August 2026 — and routes every request to the cheapest member capable of answering it. Sakana reports best overall scores on six benchmarks (Terminal Bench 2.1, GPQAD, AA-LCR, GDP.pdf, AutomationBench and its in-house SWEFish) and claims output pricing 40–60% below Sonnet 5, GPT 5.6 Terra and Kimi K3. All benchmark figures are the vendor's own and have not been independently reproduced. Because Fugu is an architecture rather than a fixed set of weights, Sakana publishes no parameter count, no context window and no training data description for it; what the customer buys is a routing policy over other companies' models, reachable through an OpenAI-compatible API.