About Sakana Fugu AI

Founded in 2026 by former Google Brain researchers David Ha and Llion Jones—a co-author of the seminal “Attention Is All You Need” paper—Sakana AI is a Japanese artificial intelligence startup redefining how frontier intelligence is accessed and applied. Our flagship product, Sakana Fugu, launched in June 2026, represents a fundamental shift away from monolithic AI models toward a dynamic, orchestrated approach.

Unlike traditional large language models, Fugu is a “model of models.” It is itself a language model trained to act as a multi-agent coordinator. When you send a request through our single, OpenAI-compatible API endpoint, Fugu intelligently decomposes the task, delegates sub-tasks to a curated pool of publicly available frontier models—including GPT-5.5, Gemini 3.1 Pro, and Claude Opus 4.8—and then synthesizes their outputs into a coherent, high-quality result.

Our Mission

We believe that access to cutting-edge AI should not be controlled by a single vendor or restricted by geopolitical boundaries. Our mission is to democratize frontier intelligence by building systems that leverage collective intelligence, mitigate vendor lock-in, and remain resilient against sudden export controls or service disruptions. We aim to provide developers, researchers, and enterprises with a reliable, cost-effective alternative that achieves top-tier performance without dependence on restricted models.

Our Values

  • Collective Intelligence: We harness the strengths of multiple expert models, combining their capabilities to produce results that rival—and often surpass—the most advanced single models available.
  • Openness and Independence: By intentionally omitting restricted models from our agent pool, we ensure our system remains accessible and compliant, reducing risks associated with vendor lock-in and geopolitical tensions.
  • Transparency and Trust: We are committed to independent validation of our performance claims. Our technical reports are publicly available, and we encourage the community to test and evaluate our system in real-world workflows.
  • Innovation Through Orchestration: We believe the future of AI lies not in bigger models, but in smarter coordination. Our architecture is designed to evolve as new models emerge, ensuring our users always have access to the best available intelligence.

Our History

Sakana AI was founded in early 2026 by David Ha and Llion Jones, two visionaries who recognized that the path to truly robust AI required moving beyond monolithic architectures. Drawing on their deep experience at Google Brain and their understanding of transformer-based systems, they set out to build an orchestration layer that could dynamically route tasks to the most suitable models.

After months of intensive research and development, Sakana Fugu was launched in June 2026 with two primary variants:

  • Fugu (Standard): Designed for everyday coding, chat, and general-purpose tasks, balancing performance with low latency.
  • Fugu Ultra: Optimized for complex, multi-step challenges such as AI research, cybersecurity analysis, and patent investigation. Fugu Ultra orchestrates a deeper pool of expert models to achieve breakthrough results.

Our technical benchmarks demonstrate that Fugu Ultra achieves performance comparable to or surpassing leading restricted models—scoring 73.7 on SWE-Bench Pro and 95.5 on GPQA-Diamond—all without using those restricted models in its agent pool. This achievement validates our core thesis: collective intelligence can outperform any single model.

Practical Utility and Availability

We understand that users searching for “Fugu AI” are often evaluating its legitimacy and practical value. Our system is currently offered as a proprietary API, with pricing structured to be competitive: Fugu Ultra costs $5 per million input tokens and $30 per million output tokens, with additional subscription plans available for high-volume users.

Early adopters have reported strong performance in research and complex reasoning tasks, though some note that response times can be slower than single-model alternatives due to the orchestration process. We are actively working to optimize latency while maintaining quality. Please note that due to GDPR compliance requirements, our service is currently restricted in the EU/EEA.

We invite you to explore Sakana Fugu and experience the power of orchestrated intelligence. Whether you are a developer seeking to reduce vendor dependency, a researcher tackling complex problems, or an enterprise looking for reliable, high-performance AI, Fugu offers a new path forward.