Sakana's Fugu Max routes models, undercuts Sonnet 5 on price

Sakana released Fugu Max on September 11, an orchestration system that routes each task to whichever model in its pool handles it best, per Sakana AI.

Fugu Max prices at $2 per million input tokens and $6 per million output tokens. Sakana says that undercuts Sonnet 5, GPT 5.6 Terra and Kimi K3 by 40-60% on output cost.

The system took the best overall score on six benchmarks, including Terminal Bench 2.1, GPQAD, AA-LCR, AutomationBench and SWEFish, and expanded the cost-performance Pareto frontier on 7 of the 10 benchmarks Sakana tested. A companion model, Fugu Ultra v2, scored 48.3 on Chartography against Claude Opus 5's 27.3 and Fable 5's 29.5, and 74.3 on DeepSWE.

What Sakana hasn't published is how the routing gets checked. Fugu's whole point is picking the best model for each task, but nothing in the release says which model handled which request. A developer paying $6 per million output tokens has no way to confirm whether a given call went to a frontier model or a cheap specialist.