Fugu Max, released by Sakana AI on September 11, 2026 alongside its flagship sibling Fugu Ultra v2, isn’t a foundation model in the usual sense. It’s a routing model trained to farm tasks out across a fixed pool of open-weight and specialized models, including NVIDIA’s Nemotron family, and to call instances of itself recursively. Sakana prices it at $2 per million input tokens and $6 per million output tokens, which the company claims is 40 to 60 percent cheaper on output than Sonnet 5, GPT-5.6 Terra, and Kimi K3.
The pitch is that orchestrating existing models well can beat training a bigger one from scratch, at a fraction of the cost. Fugu Max is the budget end of that bet, aimed at teams that want most of the capability without flagship pricing.