Skip to content
Learn · Guide · Sakana AI (Japan)

Sakana Fugu

Sakana AI's multi-agent orchestration system, delivered as a single API that assembles and coordinates many models to run complex, multi-step tasks autonomously.

Sakana AI (Japan)7 min readsakana.ai/fugu/

What is Sakana Fugu?

Sakana Fugu is a multi-agent orchestration system from Sakana AI, a Tokyo-based lab, exposed through one OpenAI-compatible API endpoint. Instead of calling a single model, you call Fugu, and it selects, switches, and coordinates a pool of expert models to complete the task.

It is built for long-running, multi-step work: users describe deploying it for autonomous research, code review, security assessments, and reproducing papers, where it operates on its own over extended periods. It ships in two variants, Fugu and the performance-optimized Fugu Ultra.

Fugu matters because it treats orchestration itself as the product: model selection and cost-versus-quality trade-offs are handled for you. The caveat for a European audience is real, so weigh it early: Fugu is not available in the EU or EEA.

Strengths

What it's best for

  • Autonomous, long-horizon tasks that benefit from coordinating several specialist models.
  • Teams that want one endpoint rather than building their own model-routing layer.
  • Research, code review, and security-assessment style work that runs unattended.
  • Controlling cost and latency without hand-tuning which model handles each step.
Limits

Where it falls short

  • Teams in the EU or EEA, where Fugu is not currently available.
  • Workloads that need a single, fixed, auditable model behind every response.
  • Buyers who want self-hostable open weights: Fugu is a hosted orchestration service.
How to use it

Ways in

Fugu is reached through an OpenAI-compatible API, so existing code that targets that shape can point at the Fugu endpoint with a base-URL and model-name change.

Pick the variant to fit the job: standard Fugu balances latency and quality, while Fugu Ultra is tuned for the hardest, highest-quality runs.

How to use it

Getting the most out of it

Give Fugu the goal and the constraints and let it decide which models to involve; the orchestration is the value, so resist micro-managing each step.

For compliance-sensitive work, use the controls over which models join the pool, and confirm data-handling terms before sending regulated data.

Pricing

What Sakana Fugu costs

Approximate, in USD, as of June 2026. Prices change often. Confirm on the official site before you rely on them.

Subscription

$20-200 / month

Includes both Fugu and Fugu Ultra, subject to usage limits. Confirm current tiers on the official site.

Pay-as-you-go (Ultra)

~$5 / 1M in, ~$30 / 1M out

Usage-based token pricing for Fugu Ultra. Confirm current rates on the official site.

Visit the official Sakana Fugu site
Try it

Example prompts

Copy these into Sakana Fugu as starting points, then adapt them to your task.

Autonomous researchCopy prompt
Investigate this question end to end: gather sources, cross-check claims, and return a cited summary with the open questions you could not resolve.
Code reviewCopy prompt
Review this pull request for correctness and security issues. Prioritize findings by severity and propose a concrete fix for each.
Routing testCopy prompt
Solve this multi-step task and, at the end, explain which kinds of sub-tasks you delegated and why, so I can judge the orchestration.
FAQ

Sakana Fugu
common questions.

Direct answers to the questions we get asked the most. If yours isn't covered, write to the team.

Work with SDEN

Putting AI into production?

We help teams choose the right models and ship them securely, self-hosted when data demands it. And we hand you the keys to run them in-house.

Sakana Fugu guide · SDEN