Sakana AI Launches Fugu Max and Fugu Ultra v2 for Cheaper, Stronger Multi-Agent Orchestration
Sakana AI's Fugu Max routes complex tasks to specialized open models at $2/$6 per 1M tokens — a new orchestration play challenging the monolithic model era.
Why it matters
Sakana AI is shipping a learned multi-model orchestration architecture that routes tasks to lean, open, specialized models (including NVIDIA Nemotron) at commodity pricing, while Fugu Ultra v2 claims peak capability on reasoning benchmarks. This is a model release and capability story — the orchestration technique and benchmark scores are the news.
The key facts
6 to knowFugu Max: $2/$6 per 1M tokens (input/output pricing)
Routes to open and specialized models, including NVIDIA Nemotron
Fugu Ultra v2: 48.3 on Chartography benchmark
Fugu Ultra v2: 74.3 on DeepSWE benchmark
Built on learned orchestration architecture
Released September 2026
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Sakana AI has released Fugu Max and Fugu Ultra v2, 2 models built on the same learned orchestration architecture. Fugu Max routes tasks to lean open and specialized models, including NVIDIA Nemotron, at $2/$6 per 1M tokens. Fugu Ultra v2 targets peak capability, scoring 48.3 on Chartography and…