Sakana Fugu: One API to Orchestrate Every AI Model

Sakana Fugu coordinates GPT, Claude, and Gemini behind a single OpenAI-compatible endpoint. Frontier-level performance through learned multi-agent orchestration — no single-vendor lock-in.

View Pricing Try Free Demos

Jun 22, 2026

Launch Date

Fugu + Ultra

Models

$20/month

Starting Price

OpenAI SDK

API Compatible

What Is Sakana Fugu?

Sakana Fugu is a multi-agent AI orchestration system built by Sakana AI, the Tokyo-based research lab known for evolutionary model merging and the AI Scientist project. Unlike ChatGPT or Claude, which route your requests to a single language model, Sakana Fugu dynamically coordinates a pool of frontier models — including GPT-5.5, Claude Opus, and Gemini 3.1 Pro — behind one OpenAI-compatible API endpoint. The result is frontier-level performance without depending on any single vendor.

The core innovation behind Sakana Fugu is learned orchestration. Traditional multi-agent systems use hand-designed workflows — if task X, call model Y. Sakana Fugu instead uses a trained 7-billion-parameter conductor model (based on two ICLR 2026 papers: TRINITY and the Conductor) that learns which models to activate, how agents should communicate, and how to combine their work. This conductor acts as a manager that directs larger, more expensive models without being one itself.

Sakana Fugu launched in general availability on June 22, 2026, after a beta period starting April 25, 2026. It ships in two variants: Sakana Fugu (balanced) for everyday coding, chat, and review workflows, and Sakana Fugu Ultra for demanding tasks like Kaggle competitions, security assessments, and multi-hour research runs. Both are accessible through a single API — you call one model, and Sakana Fugu handles the rest.

How Sakana Fugu Works

Sakana Fugu turns multi-model orchestration into a single API call. Here is what happens behind the scenes when you use Sakana Fugu:

1

You Send a Request

Call the Sakana Fugu API exactly like you'd call OpenAI. One endpoint, standard format. No SDK changes needed — if your code works with GPT, it works with Sakana Fugu.

2

Fugu Analyzes the Task

Sakana Fugu's conductor model (built on ICLR 2026 research) analyzes your request and determines which expert models to activate. It assigns Thinker, Worker, and Verifier roles dynamically.

3

Agents Coordinate

Multiple frontier models work together behind the scenes. Sakana Fugu orchestrates their communication using learned natural-language coordination strategies — not hand-designed workflows.

4

Unified Response

Sakana Fugu synthesizes all agent outputs into a single, coherent response. You receive one answer — the best result the coordinated system could produce.

Sakana Fugu Use Cases

Real-world results from Sakana Fugu users demonstrate the advantage of multi-agent orchestration over single-model approaches:

Code Review

Sakana Fugu Ultra found 20+ issues in a production codebase where competing single models found only 3. The multi-agent approach catches bugs across different categories simultaneously.

Research & Papers

Sakana Fugu reduced a 3-4 day patent landscape analysis to hours by coordinating specialized agents for search, extraction, and synthesis in parallel.

Kaggle Competitions

Sakana Fugu Ultra achieved +19.43% mean portfolio return in trading benchmarks, outperforming every single frontier model on complex multi-step optimization tasks.

Security Assessments

Sakana Fugu delivers end-to-end penetration testing assessments that stay properly scoped, with agents specializing in different vulnerability categories.

Sakana Fugu Benchmark Highlights

Sakana Fugu matches or exceeds top frontier models across engineering, science, and math benchmarks. See the full benchmark page for detailed analysis.

Benchmark Fugu Ultra GPT-5.5 Opus 4.8
SWE-Bench Pro 73.7 69.2 62.5
LiveCodeBench 93.2 88.5 85.3
GPQA-D 95.5 92.0 94.3
Humanity's Last Exam 50.0 49.8 41.4

Sakana Fugu Pricing Overview

Sakana Fugu offers subscription plans for every workload. See the full pricing page for API rates and cost comparisons.

Standard

$20/mo

Lightweight daily usage

Best for: Individual developers, casual AI tasks

Pro

$100/mo

10x usage for focused sessions

Best for: Power users, daily coding & research

Max

$200/mo

20x usage for heavy workloads

Best for: Teams, enterprises, competition-level work

Sakana Fugu FAQ

Frequently Asked Questions

What is Sakana Fugu?
Sakana Fugu is a multi-agent AI orchestration system by Sakana AI that coordinates multiple frontier language models (GPT, Claude, Gemini) behind a single OpenAI-compatible API endpoint. Instead of relying on one model, Sakana Fugu dynamically assembles and directs a pool of specialized AI agents to solve complex tasks with frontier-level performance.
How is Sakana Fugu different from ChatGPT or Claude?
ChatGPT and Claude are single-model products — you get one model per request. Sakana Fugu is fundamentally different because it orchestrates multiple models simultaneously. When you send a request to Sakana Fugu, a trained conductor model decides which experts to activate, assigns roles (Thinker, Worker, Verifier), and coordinates their outputs into one response that often surpasses any single model.
What models does Sakana Fugu use behind the scenes?
Sakana Fugu coordinates a pool of frontier models including GPT-5.5, Claude Opus, Gemini 3.1 Pro, and instances of itself recursively. The exact model routing is proprietary — Sakana Fugu's conductor model dynamically selects which agents to activate based on the specific task. With Sakana Fugu (balanced), you can opt out specific agents for compliance reasons.
Is Sakana Fugu available in the EU?
No, Sakana Fugu is currently unavailable in EU/EEA member states. Sakana AI states that GDPR compliance is in progress. Users in the EU cannot access Sakana Fugu's API or subscription plans at this time. Check the official Sakana Fugu console for updates on regional availability.
How much does Sakana Fugu cost?
Sakana Fugu offers three subscription tiers: Standard at $20/month for lightweight usage, Pro at $100/month for 10x usage, and Max at $200/month for 20x heavy workloads. Sakana Fugu also offers pay-as-you-go API pricing: Fugu Ultra costs $5/1M input tokens and $30/1M output tokens. Subscribe before July 2026 for one free month.
View all Sakana Fugu frequently asked questions →

Ready to Try Sakana Fugu?

Explore Sakana AI's technology with free interactive demos, or dive into the full Sakana Fugu pricing and integration guide.

Try Free Demos API Setup Guide