- Published on
Sakana AI Launches Fugu Max and Fugu Ultra v2 for Cheaper, Stronger Multi-Agent Orchestration
- Authors
- Name
- DeepHorus
September 10, 2026, Sakana AI launched Fugu Max and Fugu Ultra v2. They target cheaper and stronger multi-agent orchestration. Both are live today through Sakana's OpenAI-compatible API. No open weights are available; the service excludes the EU/EEA.
Fugu is not a single foundation model. It is a learned orchestrator that routes tasks across a model pool behind one API. Fugu Max optimizes for output per dollar. Fugu Ultra v2 targets capability on hard, multi-step tasks.
Fugu entered beta in April, GA in June, and added Fugu-Cyber in July. The system builds on two ICLR 2026 papers, TRINITY and the Conductor. Training combines fine-tuning, evolutionary algorithms, and reinforcement learning.
Fugu Max adds open-weights and specialized models, including NVIDIA's Nemotron family. Pricing: $2 per million input tokens and $6 per million output tokens. Output prices are 40% to 60% lower than Sonnet 5, GPT 5.6 Terra, and Kimi K3. It tops six benchmarks.
Fugu Ultra v2 targets complex reasoning, autonomous research, and full-stack software development. Its largest gains appear in sustained reasoning over visual and structured data. Chartography shows a score of 48.3.
Sakana frames the release as a two-axis problem of capability and cost. Sending a simple lookup to a huge model wastes money. Fugu picks the cheapest machinery that can solve the task. The company describes this with the Pareto frontier.
The implications are significant for enterprises seeking lower-cost frontier results. Fugu Max expands the cost-performance frontier on seven of ten benchmarks. It puts elite models within striking distance at two to six times lower cost. But no open weights and EU exclusion may limit adoption.
Looking ahead, Sakana's fast cadence suggests more iterations are coming. Developers can test both models through the hosted API today. The future will likely bring a wider model pool and refined orchestration strategies.
