Fugu-Ultra v1.1 Release Announcement 🐡

Sakana AI has released Fugu-Ultra v1.1. This update integrates the latest frontier models to improve performance across all benchmarks, showing gains of up to 7.9 points compared to v1.0. It has demonstrated very strong performance on ProgramBench and Terminal Bench 2.1.

Fugu-Ultra v1.1 has further strengthened its coding, agentic task, and advanced reasoning capabilities, and is offered at the same price as the existing v1.0. Fugu is a full multi-agent orchestration system accessible through a single model API.

The core working principles of Sakana Fugu are as follows:

• Orchestration: Fugu itself acts as an LLM, calling and managing LLMs within a diverse pool of agents.
• Automated Management: It automatically handles model selection, task delegation, verification, and synthesis, and forms teams of expert models to solve complex problems.
• Model Lineup: It is divided into Fugu, for everyday tasks and low latency, and Fugu Ultra, the flagship model for high-difficulty tasks such as AI research and cybersecurity analysis.

Additionally, Sakana AI is realizing collective AI intelligence by introducing AB-MCTS, a new inference-time scaling algorithm that enables multiple frontier models (Gemini 2.5 Pro, o4-mini, DeepSeek-R1, etc.) to collaborate.