About Fugu AI – The Model of Models
Fugu AI, officially known as Sakana Fugu, is a groundbreaking AI orchestration system developed by the Japanese startup Sakana AI. Co-founded by former Google Brain researchers David Ha and Llion Jones—a co-author of the foundational “Attention Is All You Need” paper—Sakana AI has reimagined how artificial intelligence can work. Instead of relying on a single large language model, Fugu acts as a “model of models,” dynamically delegating sub-tasks to a pool of publicly available frontier models like GPT-5.5, Gemini 3.1 Pro, and Claude Opus 4.8. It then synthesizes their outputs through a single, OpenAI-compatible API endpoint. Launched in June 2026, Fugu AI offers two variants: the standard Fugu for everyday coding and chat tasks, and the flagship Fugu Ultra for complex, multi-step challenges such as AI research, cybersecurity analysis, and patent investigation.
Our Mission
Our mission is to democratize access to frontier AI performance by breaking free from the limitations of monolithic models. We believe that collective intelligence—orchestrating the strengths of multiple expert models—can achieve results that rival or surpass even the most restricted systems. By providing a resilient, vendor-independent alternative, we aim to empower developers, researchers, and enterprises to innovate without fear of sudden access cuts due to geopolitical export controls or vendor lock-in.
Our Values
- Innovation Through Orchestration: We champion a non-monolithic approach, using a trained language model as a multi-agent coordinator to route tasks intelligently.
- Resilience and Independence: Our system is designed to mitigate risks from export restrictions and single-vendor dependencies, ensuring consistent access to top-tier AI capabilities.
- Transparency and Practicality: We provide clear pricing—Fugu Ultra costs $5 per million input tokens and $30 per million output tokens—and offer subscription plans to suit diverse workflows.
- Continuous Improvement: We actively validate performance through benchmarks like SWE-Bench Pro (73.7) and GPQA-Diamond (95.5), and we welcome user feedback to refine our orchestration.
Our History
Sakana AI was founded in 2023 by David Ha and Llion Jones, bringing together deep expertise in neural networks and transformer architectures. The team recognized that the AI landscape was becoming increasingly fragmented, with powerful models locked behind proprietary walls or restricted by export controls. In response, they developed Fugu—a system that leverages the collective intelligence of multiple frontier models without relying on any single one. After extensive research and development, Fugu AI was officially launched in June 2026. The standard Fugu variant quickly gained traction for everyday tasks, while Fugu Ultra demonstrated competitive benchmark scores, achieving performance comparable to leading restricted models like Anthropic’s Claude Mythos Preview and Fable 5—all without using those restricted models in its agent pool. Today, Fugu AI is available as a proprietary API, with restrictions in the EU/EEA due to GDPR compliance, and continues to evolve based on user insights and independent validation.
Why Choose Fugu AI?
Users turn to Fugu AI to investigate a new, non-monolithic path to frontier AI performance. Our orchestration architecture offers a practical solution to vendor lock-in and geopolitical risks, while our pricing remains competitive for both development and enterprise workflows. Whether you are tackling complex research, automating cybersecurity analysis, or building next-generation applications, Fugu AI provides a resilient, intelligent, and cost-effective alternative.