Sakana AI ships two new Fugu models for multi-agent orchestration
single source · 1 articles · release · confidence: high · first seen 2026-09-11 06:34 UTC
Sakana AI has released Fugu Max and Fugu Ultra v2, two models for coordinating multiple AI agents. Both use the same learned orchestration architecture, the company says. Fugu Max is the lower-cost option: it routes tasks to lean open and specialised models, including NVIDIA Nemotron, and is priced at $2/$6 per 1M tokens. Fugu Ultra v2 is the higher-capability option, reporting scores of 48.3 on Chartography and 74.3 on DeepSWE, without evaluation dates or harness details. The release gives developers a cheaper routing model and a more capable one.
What this means for you
If you run multi-agent workloads, test Fugu Max's $2/$6 per 1M token pricing against your current router. Don't benchmark against Fugu Ultra v2's 48.3 Chartography or 74.3 DeepSWE until the evaluation dates and harness are published.
Key facts
- ·Sakana AI released Fugu Max and Fugu Ultra v2, two models for multi-agent orchestration. source
- ·Fugu Max is priced at $2/$6 per 1M tokens. source
- ·Fugu Max routes tasks to lean open and specialised models, including NVIDIA Nemotron. source
- ·Fugu Ultra v2 has reported scores of 48.3 on Chartography and 74.3 on DeepSWE; the article does not include evaluation dates or harness details. source
What the sources say
- MarkTechPost — Announcement-style post listing the two models, prices, and benchmark figures.
Sources
The original reporting. Follow these — they did the work.