Sakana Fugu Ultra is part of a family of orchestrator models that coordinate pools of existing frontier language models rather than acting as a single trained base system. According to the Sakana Fugu Technical Report, these orchestrator models are themselves language models trained to understand user queries and dynamically devise agentic scaffolds, accessing performance beyond any individual LLM agent through adaptive routing and multi-agent coordination. Fugu Ultra is positioned as the quality-first variant within that family, coordinating a deeper pool of expert agents and routing between one to three agents depending on the problem, with a higher cost profile reflecting that depth.
The intended use case for Fugu Ultra is hard, high-stakes work where answer quality matters more than latency, including complex multi-step engineering, scientific, and reasoning problems. A third-party report cites a SWE-Bench Pro score of 73.7% for Fugu Ultra, placing it ahead of Claude Opus 4.8 at 69.2%, though those numbers come from the provider rather than independent reproduction. The model is exposed through an OpenAI-compatible API and is paired with a lighter Fugu variant for everyday latency-sensitive work, giving teams a way to trade speed for depth on the same orchestration platform when tackling the most demanding agentic tasks.
