Tool·AI Models & Capabilities·Added today

Sakana Fugu

Also known as: Fugu, Fugu Ultra, Fugu Max, Sakana AI Fugu

A multi-model orchestration system from Sakana AI that routes tasks across multiple frontier models rather than relying on a single one. It exposes a single API but internally assembles a workflow of specialized models to handle different parts of a request.

Sakana AI, a Tokyo-based research lab, developed Fugu as an orchestration layer rather than a single model. The premise: no one frontier model is best at everything, so Fugu acts as a coordinator that breaks a task into components, assigns each to a model that specializes in it, and fuses the results. The 'single API, many models under the hood' architecture makes it straightforward to drop into an existing agent stack without changing the calling code.

Sakana released Fugu Max v1.0 and Fugu Ultra v2.0 on September 11, 2026. In benchmark testing, Fugu Ultra scored 74.3 on DeepSWE (a software engineering benchmark) and 48.3 on Chartography, at $5 per million input tokens. Sakana's argument is that for enterprises blocked from accessing certain frontier models by export controls or regulatory constraints, Fugu provides a sovereignty-friendly alternative that can match or beat frontier performance by combining smaller, accessible models.

Fugu is Specialist-level vocabulary today, relevant mainly to builders evaluating agent harnesses and cost optimization strategies. It's notable as an early commercial example of the orchestrate-don't-just-route pattern, the same shift GitHub's HydraFusion represents. Both treat 'which workflow should I run?' as the core question, not 'which single model should I call?' If that pattern continues to spread, the term 'multi-model orchestrator' may become as common in 2027 as 'model router' is today.

This definition is AI-generated and refreshed weekly. It may contain inaccuracies. Use your own judgment, especially for production decisions.

Related terms