Japanese AI company Sakana AI has introduced Fugu Max and Fugu Ultra v2, two new products built around the same core idea: a model orchestrator that reviews a task, assigns different models to handle parts of it, lets them cross-check one another, and then combines the final answer. According to the brief, the first-generation Fugu used this approach to assemble teams of models including GPT, Claude, and Gemini, reaching performance close to, and in some cases above, the then-stronger Fable 5.
The new lineup splits into two clear directions. Fugu Max is positioned as the lower-cost option, expanding its model pool with more open-weight and specialist models while favoring cheaper models that are still sufficient for the job. Sakana said that when performance is close to top-tier models, the cost is about 1/2 to 1/6 of those alternatives. Fugu Ultra v2, by contrast, is aimed strictly at top-end performance. Sakana said it ranked first or tied for first in 5 of 8 benchmarks, even though its agent pool does not include Fable 5, Fable 5.1, or GPT-6 Astra. The company did not disclose Max’s cache hit rate or total token cost per task.
Japanese AI company Sakana AI has released Fugu Max and Fugu Ultra v2. Both are essentially model orchestrators: they first inspect a task, then assign the right models to different parts of the work, have those models check one another, and finally synthesize an answer.
Two product paths for the Fugu lineup
According to the brief, the first-generation Fugu used this multi-model setup to assemble teams built from models such as GPT, Claude, and Gemini, reaching performance close to, and in some cases above, the then-stronger Fable 5.
This time, Sakana split the lineup into two tracks. Fugu Max is focused on lowering costs. Its model pool adds more open-weight and specialist models, and it prioritizes models that are cheaper while still good enough for the task. The company said that when performance is close to top-tier models, the cost is about 1/2 to 1/6 of those alternatives.
Ultra v2 is aimed at peak performance
Fugu Ultra v2 takes the opposite approach and focuses only on maximum performance. Sakana said it placed first or tied for first in 5 of 8 benchmarks, and that its agent pool does not include Fable 5, Fable 5.1, or GPT-6 Astra.
Trade-offs and missing cost details
The brief also notes a drawback: switching back and forth across multiple models can reduce context-cache reuse and add extra orchestration tokens. Sakana did not disclose Fugu Max’s cache hit rate or the total token cost for a single task.
It also said users had criticized the original Fugu for being slow and burning through usage quotas quickly.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
