Skip to content
Sakana AI

Fugu Ultra

An AI orchestration system ('conductor') that dynamically routes tasks to multiple frontier models.

Quality

Modality

text

Context

1M tokens

Access

closed

Fabian's Take

FM

"Fugu isn't a traditional model — it's somewhere between a tool and a model. It acts as a 'conductor,' deciding which model (Claude, GPT, or an open-weights model) is best suited for your prompt and routing it there dynamically. I choose it as the engine to power my agents when I want the absolute best result without worrying about which API to call."

Fugu Ultra, developed by Tokyo-based Sakana AI, is a unique entry in the AI landscape. It is not a traditional Large Language Model; rather, it is an AI orchestration system, often described as a “conductor.”

The orchestrator approach

When you send a prompt to Fugu, it analyzes the complexity and domain of the request, then dynamically routes it to the most suitable frontier model available (such as GPT-5.5, Claude Opus, Gemini Pro, or various open-source models). This allows it to achieve exceptionally high benchmark scores across a wide range of tasks, as it leverages the specific strengths of multiple models while masking the routing complexity from the user.

The Verdict

Best for: Complex tasks requiring the combined strengths of multiple frontier models.

Pros

  • Consistently high benchmark scores across multiple domains
  • Avoids vendor lock-in by routing to the best model for the job
  • Focuses on 'AI sovereignty'

Cons

  • Acts more like a router than a traditional LLM
  • Pricing increases significantly for very large contexts

Specs

  • Pricing $5/M input, $30/M output
  • Cost Tier premium
  • Speed Tier moderate
  • License Proprietary
Developer Docs