Sakana AI, a Tokyo-based artificial intelligence company, announced Fugu on June 22, 2026, a new multi-agent system designed to dynamically orchestrate the world’s best models for complex, multi-step tasks.
Fugu itself is a language model trained to understand delegation, inter-agent communication, and result aggregation. The system routes requests to specialized models based on task requirements, aggregating results into unified responses through a single API endpoint.
The approach builds on Sakana AI’s research in learned model orchestration, including papers presented at ICLR 2026 on Trinity and The Conductor. Rather than relying on a single large model for all tasks, Fugu acts as an intelligent routing layer that matches work to optimal performers.
Sakana offers two model variants. Fugu balances performance and latency for everyday tasks, while Fugu Ultra targets maximum accuracy on demanding, multi-step problems requiring careful reasoning and synthesis.
The company claims Fugu Ultra rivals performance benchmarks of competitors like Anthropic’s Fable 5 and Mythos Preview without the geopolitical risks associated with export controls on frontier AI models.
Both models are accessible via an OpenAI-compatible API, making integration straightforward for developers already familiar with standard model interfaces. Pricing starts at $20 monthly for the Standard plan, $100 for Pro, and $200 for Max tier access.
Sakana AI’s approach addresses a critical challenge in AI deployment: complex tasks often require different models at different stages. By automating model selection and orchestration, Fugu reduces deployment friction for enterprise teams building multi-step AI workflows.
The announcement positions Sakana AI as a key player in the emerging orchestration layer of AI infrastructure, where routing and coordination become as important as individual model performance.
