Sakana AI

Sakana AI is a Tokyo-based artificial intelligence company founded in 2023 by David Ha, Llion Jones, and Ren Ito, focusing on nature-inspired algorithms like evolution and collective intelligence to develop AI models.

Sakana AI Co, Ltd. is a Japanese artificial intelligence company headquartered in Tokyo. The company focuses on research in evolution and collective intelligence of AI, drawing inspiration from natural systems. Its name derives from the Japanese word for fish, symbolizing how a school of fish forms a coherent entity from simple individual rules - an analogy for collective intelligence.

The company was founded by David Ha, Llion Jones, and Ren Ito. Llion Jones co-authored the seminal 2017 paper "Attention Is All You Need" while working at Google, which introduced the Transformer (architecture) architecture that underpins modern Large language models. Sakana AI raised $30 million in seed funding from Lux Capital and Khosla Ventures, followed by approximately $200 million in a Series A round in 2024 from investors including Mitsubishi UFJ, SMBC, Mizuho, Itochu, KDDI, Nomura, and Nvidia. The Nikkei estimated the company's value at 19 billion yen in 2024.

Founding and Mission

Sakana AI was established in Tokyo with the goal of democratizing AI development through nature-inspired computational methods. The founders sought to move beyond the paradigm of scaling up monolithic Neural networks, instead exploring how simple interacting components can produce complex, adaptive behaviors. The company's research agenda is rooted in evolutionary computation and swarm intelligence, areas that have historically been less prominent than deep learning but offer alternative pathways to creating capable AI systems.

David Ha previously worked at Google Brain, where he contributed to research on neural network architectures and reinforcement learning. Llion Jones brought expertise from his work on the transformer architecture, which has become foundational to modern Generative AI. Ren Ito, who serves as CEO, provided entrepreneurial and strategic leadership. The trio's combined background positioned Sakana AI to bridge cutting-edge machine learning research with practical product development.

Nature-Inspired Algorithms

Sakana AI's core research areas are evolution and collective intelligence. The company develops algorithms that mimic biological processes such as natural selection, mutation, and recombination to optimize AI models. This approach contrasts with traditional methods that rely on manually designed architectures and fixed training procedures. By treating models as organisms that can be bred and evolved, Sakana AI aims to discover novel solutions that might be missed by human intuition.

Collective intelligence is another central theme. The company studies how multiple AI agents can coordinate and share information to solve problems more effectively than any single agent. This draws on principles observed in fish schools, bird flocks, and insect colonies, where simple local rules give rise to sophisticated group behavior. Sakana AI applies these ideas to multi-agent systems, potentially enabling more robust and scalable AI deployments.

Model Breeding Method

In January 2024, Sakana AI developed a method to build new AI models by 'breeding' multiple existing models. This process involves combining and mutating components from different models, analogous to genetic crossover in biological evolution. The technique allows for the creation of models with improved capabilities without requiring massive computational resources for training from scratch. Sakana AI sees this as a means to democratize AI development, as smaller organizations and researchers can participate in model creation without access to large-scale computing infrastructure.

The breeding method leverages insights from evolutionary computation and Model Pruning to selectively combine beneficial traits. It can operate on open-source models, which are often released with permissive licenses, enabling a collaborative ecosystem of model improvement. This approach has implications for reducing the environmental and financial costs associated with training large models, making AI research more accessible globally.

AI Scientist Project

Sakana AI is also developing a model called the AI Scientist, which automates the entire process of scientific research. This ambitious project aims to create an AI system that can generate hypotheses, design experiments, run simulations, analyze results, and write papers. The AI Scientist could accelerate discovery in fields ranging from biology to materials science by handling routine aspects of the scientific method.

The project integrates multiple AI capabilities, including Large language models for reasoning and text generation, as well as specialized modules for data analysis and experimental design. While still in early stages, the AI Scientist represents a significant step toward autonomous research systems. Sakana AI positions this work as complementary to human scientists, potentially augmenting their productivity rather than replacing them.

Funding and Investors

Sakana AI's financial backing reflects strong confidence from both venture capital and corporate investors. The seed round of $30 million was led by Lux Capital and Khosla Ventures, two prominent Silicon Valley firms known for backing frontier technology startups. The Series A round, announced in 2024, brought in approximately $200 million from a consortium of Japanese financial institutions and technology companies, including Mitsubishi UFJ, SMBC, Mizuho, Itochu, KDDI, Nomura, and Nvidia.

The involvement of Nvidia, a leading supplier of Artificial intelligence hardware, suggests potential synergies in optimizing Sakana AI's algorithms for specific chip architectures. The participation of major Japanese banks and trading companies indicates strategic interest in AI's application to finance, logistics, and telecommunications. This diverse investor base provides Sakana AI with both capital and access to industry partnerships.

Position in the AI Landscape

Sakana AI operates in a competitive field dominated by large labs such as OpenAI, Anthropic, and Google DeepMind. However, its focus on nature-inspired algorithms differentiates it from these peers, which primarily pursue scaling of existing architectures. The company's Tokyo location also positions it within Japan's growing AI ecosystem, alongside efforts from Sony AI and Fujitsu. Sakana AI's approach could complement the work of academic institutions like MIT CSAIL and Stanford AI Lab, which explore similar themes in evolutionary computation.

The company's emphasis on efficiency and democratization resonates with broader trends in the AI community, such as the development of smaller, more specialized models. By reducing the computational barriers to model creation, Sakana AI may enable a more distributed and inclusive AI research landscape. Its methods could also inform the design of Residual Network (ResNet)s and other architectures that benefit from evolutionary optimization.

Future Directions

Looking ahead, Sakana AI plans to expand its research into new applications of collective intelligence, including robotics and multi-agent systems. The company is also exploring partnerships with academic and industrial collaborators to validate its methods in real-world settings. As of 2024, Sakana AI continues to refine its breeding techniques and the AI Scientist, with the goal of releasing tools that researchers can use to accelerate their own work.

The company's success will depend on whether its nature-inspired approaches can achieve results competitive with traditional deep learning at scale. Early results are promising, but broader adoption will require demonstrating clear advantages in efficiency, robustness, or capability. Sakana AI's unique position at the intersection of evolution, collective intelligence, and practical AI development makes it a notable player to watch in the evolving landscape of Machine learning.