Sakana AI was founded in Tokyo in 2023 by David Ha, former research director at Google Brain, and Llion Jones, one of the original co-authors of the seminal “Attention Is All You Need” paper that introduced the Transformer architecture. The company’s name references the Japanese word for fish, reflecting its research philosophy inspired by the collective intelligence of natural systems — the way schools of fish or flocks of birds exhibit intelligent group behavior from simple individual rules.
Research Philosophy
Sakana pursues “nature-inspired AI” — developing models and training approaches that draw on evolutionary biology, swarm intelligence, and emergence rather than pure scale-based approaches. The company is particularly interested in model merging (combining existing trained models rather than training from scratch) and evolutionary search methods that can automatically discover better architectures and training recipes. This approach is partly pragmatic: by not competing on raw compute, Sakana can do original research with a smaller team.
Notable Research
Sakana’s most viral research release was “The AI Scientist,” a system that can autonomously conduct machine learning research — proposing hypotheses, writing code, running experiments, analyzing results, and drafting papers. While the system has limitations, it demonstrated autonomous scientific reasoning capabilities that attracted significant attention from the research community. Sakana also published work on Evolutionary Model Merge, showing that combining model weights from different trained models could produce better results than either model alone.
Competitive Position
Sakana occupies a unique niche as an AI research lab in Japan, a country with historically significant AI research but relatively few frontier AI startups. The company benefits from Japan’s government interest in building domestic AI capability, as well as strong Japanese corporate backing from Sony and others. Internationally, Sakana differentiates by focusing on research problems that larger labs deprioritize — efficiency, composability, and evolutionary optimization rather than raw benchmark performance.
Outlook
Sakana closed a major funding round in late 2024, giving it the runway to pursue long-horizon research. The company’s commercial path involves licensing its research, building enterprise applications in Japan, and potentially deploying its model merging and efficiency research in products. Japan’s government support for AI and the country’s large technology sector provide a distinctive home market. Sakana’s trajectory will depend on whether its alternative research paradigm produces capabilities that matter — or whether scale-first approaches continue to dominate.