1
AutoGen
Facilitates conversational multi-agent workflows where agents communicate asynchronously to achieve complex tasks.
AI Tool
Fugu Review
Fugu is a multi-agent orchestration system functioning as a conductor LLM, trained to dynamically route incoming tasks to the optimal agent from a swappable pool of other LLMs.
shipped Jun 23, 2026aifreemium
1Fugu and Fugu Ultra were officially launched as generally available (GA) products on June 22, 2026.
2Approximately 500 beta users tested the Fugu system prior to its general availability.
3Fugu Ultra scored 93.2 on LiveCodeBench, outperforming Fable 5's 89.8 in coding performance.
4Subscription plans for Fugu start at $20 per month for the Standard tier.
Fugu at a Glance
Similar Tools
Compare Alternatives
Other tools you might consider
View on Stork→
2
CrewAI
Provides an open-source framework for building and orchestrating collaborative AI agents with advanced memory management and checkpointing capabilities.
3
LiteLLM
Offers a unified API interface to over 100 LLM providers with features like fallback, load balancing, and budget tracking.
4
RouteLLM
A principled open-source framework for dynamically selecting the most cost-effective LLM for each query based on complexity and performance.
overview
What is Fugu?
Fugu is a multi-agent orchestration system tool developed by Sakana AI that enables corporations, financial institutions, and think tanks to orchestrate diverse models for specific tasks. It functions as a conductor LLM, dynamically routing incoming tasks to the optimal agent from a swappable pool of other LLMs. Launched on June 22, 2026, Fugu aims to provide frontier-level AI capabilities while mitigating the risks associated with relying on a single AI provider. The system presents a coordinated pool of specialized AI models through a single OpenAI-compatible API. Fugu itself is a language model trained to dynamically select, delegate tasks to, and synthesize responses from various underlying LLMs, including commercial models like Google Gemini 3.1 Pro, OpenAI GPT-5.5, and Anthropic Claude Opus 4.8, as well as Sakana AI's own models. It is available in two variants: Fugu, optimized for strong performance and low latency for everyday tasks such as coding and powering chatbots; and Fugu Ultra, designed for maximum answer quality on complex, multi-step problems, suitable for AI research and cybersecurity analysis.
quick facts
Quick Facts
| Attribute | Value |
|---|---|
| Developer | Sakana AI |
| Business Model | Hybrid (Freemium, Subscription SaaS, Usage-based) |
| Pricing | Freemium starting at $20/month (Standard subscription) or $5 per 1 million input tokens (Pay-As-You-Go) |
| Platforms | API |
| API Available | Yes (OpenAI-compatible) |
| Integrations | Google Gemini 3.1 Pro, OpenAI GPT-5.5, Anthropic Claude Opus 4.8, Sakana AI models |
| Launched | June 22, 2026 |
| HQ | Tokyo, Japan |
features
Key Features of Fugu
Fugu is engineered as a sophisticated multi-agent orchestration system, leveraging a conductor LLM to manage and optimize AI workflows. Its core functionality revolves around intelligent task routing and the flexible integration of diverse language models.
- 1Multi-agent orchestration system for complex AI workflows.
- 2Functions as a conductor LLM, dynamically routing incoming tasks.
- 3Routes tasks to the optimal agent from a swappable pool of other LLMs.
- 4Utilizes a diverse agent pool including Google Gemini 3.1 Pro, OpenAI GPT-5.5, and Anthropic Claude Opus 4.8.
- 5Provides a single OpenAI-compatible API for accessing coordinated AI models.
- 6Offers two variants: Fugu for performance/latency and Fugu Ultra for maximum answer quality.
- 7Mitigates vendor lock-in and geopolitical risks by reducing reliance on single AI providers.
- 8Employs a 'passthrough billing' model, charging a single rate based on the top-tier model used plus a Sakana margin.
use cases
Who Should Use Fugu?
Fugu is designed for organizations requiring advanced AI capabilities, particularly those seeking to optimize performance, ensure data sovereignty, and manage complex, multi-step tasks across various domains. Its architecture supports a range of demanding applications.
- 1Corporations and Financial Institutions: For generating deep, long-horizon strategy reports and executive slides.
- 2Think Tanks and Research Organizations: For automating scientific discovery, reproducing scientific papers, and conducting literature and patent investigations.
- 3Organizations with strict data governance: For providing AI sovereignty and mitigating vendor lock-in through orchestrating diverse models.
- 4Engineering teams and Data science units: For coding, code review, automating cybersecurity assessment cycles, and penetration testing.
- 5Japanese Enterprises: For specialized tasks such as Japanese handwriting analysis and other region-specific applications.
pricing
Fugu Pricing & Plans
Sakana AI offers two primary pricing structures for Fugu: subscription plans for individual users and everyday use, and a Pay-As-You-Go model for corporate clients and heavy production workloads. All subscription tiers include access to both Fugu and Fugu Ultra models. The 'passthrough billing' model ensures that even with multiple agents, fees are not stacked, but a single rate is charged based on the top-tier model utilized, plus a Sakana margin.
- 1Standard Subscription: $20 per month.
- 2Pro Subscription: $100 per month (10x Standard usage volume).
- 3Max Subscription: $200 per month (20x Standard usage volume).
- 4Pay-As-You-Go Input Tokens: $5 per 1 million tokens (for contexts up to 272K tokens), increasing to $10 per 1 million tokens for contexts exceeding 272K tokens.
- 5Pay-As-You-Go Output Tokens: $30 per 1 million tokens (for contexts up to 272K tokens), increasing to $45 per 1 million tokens for contexts exceeding 272K tokens.
- 6Pay-As-You-Go Cached Input Tokens: $0.50 per 1 million tokens (for contexts up to 272K tokens), increasing to $1.00 per 1 million tokens for contexts exceeding 272K tokens.
competitors
Fugu vs Competitors
Fugu is strategically positioned to reduce dependence on any single AI provider, offering a hedge against vendor lock-in and geopolitical risks. Sakana AI claims Fugu Ultra performs comparably to leading models such as Anthropic's Fable 5 and Mythos Preview across key engineering, scientific, and reasoning benchmarks. In internal benchmark tests, Fugu models reportedly outperformed Google Gemini 3.1 Pro, OpenAI GPT-5.5, and Anthropic Claude Opus 4.8 in tasks including automated research and financial forecasting. For instance, Fugu Ultra scored 93.2 on LiveCodeBench, exceeding Fable 5's 89.8, and 95.5 on GPQA-D, surpassing Mythos Preview's 94.6. However, Fugu Ultra reportedly fell short of Fable 5 on 'Humanity's Last Exam' (50.0 vs 53.3) and lagged behind GPT-5.5 and Opus 4.8 on long-context recall and cybersecurity benchmarks, respectively.
1
Facilitates conversational multi-agent workflows where agents communicate asynchronously to achieve complex tasks.
Similar to Fugu in orchestrating multiple LLMs/agents, AutoGen emphasizes a chat-centric, conversational model for agent interaction, providing a flexible framework for developers. Fugu is described as a 'conductor LLM' for routing, while AutoGen focuses on the collaborative conversational aspect of agents.
2
Provides an open-source framework for building and orchestrating collaborative AI agents with advanced memory management and checkpointing capabilities.
Like Fugu, CrewAI focuses on multi-agent orchestration and task execution. CrewAI offers sophisticated memory and checkpointing for production-ready agents, whereas Fugu highlights its 'conductor LLM' for dynamic routing.
3
LiteLLM↗
Offers a unified API interface to over 100 LLM providers with features like fallback, load balancing, and budget tracking.
LiteLLM acts primarily as an LLM router and gateway, which is a core component of Fugu's dynamic routing to optimal LLMs. While Fugu focuses on orchestrating agents, LiteLLM directly manages and optimizes calls to various LLM providers, offering cost optimization through intelligent routing.
4
A principled open-source framework for dynamically selecting the most cost-effective LLM for each query based on complexity and performance.
RouteLLM directly competes with Fugu's core function of dynamically routing incoming tasks to the optimal LLM by specializing in cost-effective LLM selection. Fugu's scope appears broader to multi-agent orchestration, while RouteLLM is more focused on the intelligent routing of individual LLM queries.
❓
Frequently Asked Questions
+What is Fugu?
Fugu is a multi-agent orchestration system tool developed by Sakana AI that enables corporations, financial institutions, and think tanks to orchestrate diverse models for specific tasks. It functions as a conductor LLM, dynamically routing incoming tasks to the optimal agent from a swappable pool of other LLMs.
+Is Fugu free?
Fugu operates on a freemium model. While a free tier is available, detailed pricing includes subscription plans starting at $20 per month for the Standard tier, and a Pay-As-You-Go model with input tokens priced at $5 per 1 million tokens for contexts up to 272K tokens.
+What are the main features of Fugu?
Key features of Fugu include its multi-agent orchestration system, functioning as a conductor LLM for dynamic task routing, utilization of a swappable pool of other LLMs (e.g., Google Gemini 3.1 Pro, OpenAI GPT-5.5), provision of a single OpenAI-compatible API, and offering two variants: Fugu for performance and Fugu Ultra for maximum answer quality. It also mitigates vendor lock-in and uses a 'passthrough billing' model.
+Who should use Fugu?
Fugu is intended for corporations, financial institutions, think tanks, and organizations with strict data governance requirements. It is also suitable for engineering teams and data science units, particularly for tasks such as automating scientific discovery, generating strategy reports, cybersecurity assessments, and mitigating vendor lock-in.
+How does Fugu compare to alternatives?
Fugu differentiates itself from competitors like AutoGen, CrewAI, LiteLLM, and RouteLLM by focusing on a 'conductor LLM' for dynamic routing within a multi-agent orchestration system. While AutoGen emphasizes conversational workflows and CrewAI offers advanced memory management, Fugu's strength lies in its intelligent delegation to a swappable pool of diverse LLMs to provide frontier-level AI capabilities and reduce reliance on single providers.
More on Stork
Related AI Tools
Other tools in this category, ranked by community signal
Code Rabbit
🤖 AI Tools
An AI-powered platform for automated code reviews, planning, and development workflows, integrating with Git platforms to provide real-time feedback and suggestions.
G
GLM-5.2
🤖 AI Tools
A 750 billion parameter, open-source large language model from Zhipu AI, designed for coding tasks with a focus on cost-effectiveness and long-horizon task execution.
Kimi K2.7 Code
🤖 AI Tools
Kimi K2.7 Code is Moonshot AI's coding-focused agentic model, built with a Mixture-of-Experts architecture for improved long-horizon coding tasks and token efficiency.
Walrus Memory
🤖 AI Tools
Walrus Memory is a decentralized, universal memory layer for AI agents that enables persistent context sharing across different AI tools.
Sorce
🤖 AI Tools
Sorce is an AI-powered job search platform that simplifies the application process by allowing users to swipe right on job listings, after which the platform's AI agent handles the application submission.
SubQ
🤖 AI Tools
SubQ is a Large Language Model (LLM) built on a sub-quadratic sparse attention architecture designed for extreme efficiency and performance on very long context tasks.
For builders
This page is doing a job for someone else’s tool.
AI agents read it. Buyers find it. Backlinks accrue. Your tool can have one too — live in 24 hours, indexed by Claude, ChatGPT, and Perplexity, queryable via MCP.