According to CoinMeta, Sakana AI has released Fugu Max and Fugu Ultra v2. Both of these products are essentially "model schedulers" that can select appropriate models based on tasks for division of labor and checking, ultimately synthesizing answers. Fugu Max is designed to save costs; its model pool includes more open-weight and professional models, with costs ranging from 1/2 to 1/6 of those of top-tier models. On the other hand, Fugu Ultra v2 aims for the highest performance, achieving first place or tied first place in 5 out of 8 evaluations. Nevertheless, switching between multiple models may reduce the reuse of context cache and incur additional orchestration token. Sakana has not yet announced the cache hit rate of Max or the total token cost per task.