Sakana AI was co-founded by a co-author of the transformer paper — literally one of the people who invented the architecture that runs ChatGPT. He and David Ha left Google to build AI in Tokyo using evolutionary algorithms and "nature-inspired" approaches instead of brute-force scaling. The "AI Scientist" system they released can run experiments and write papers autonomously. Whether small and evolutionary beats large and brute-force is the most interesting open question in AI research right now.
Founded
2023
HQ
Tokyo, Japan
Total Raised
$54M raised
Founder
David Ha, Llion Jones
Status
Active — publishing research, developing commercial AI products
Website
sakana.aiTHE ORIGIN STORY
David Ha was the head of research at Google Brain Japan and one of the leading researchers in evolutionary algorithms and neural architecture. Llion Jones was a co-author of the original "Attention Is All You Need" paper — the 2017 paper that introduced the transformer architecture that underlies every modern AI system from GPT to Gemini.
In 2023 both left Google to found Sakana AI in Tokyo with a thesis: the current trend of ever-larger AI models is not the only path. Inspired by biological evolution and the collective intelligence of fish schools (sakana means "fish" in Japanese), they are building small, diverse, specialized AI models that can combine and collaborate rather than training single massive models.
WHAT THEY ACTUALLY DO
Sakana AI earns revenue through research partnerships and licensing of its models and techniques to enterprise customers. It also sells API access to its deployed models.
The company is at an early stage of commercialization and remains primarily R&D focused, with revenue expected to scale as models are productized.
THE PRODUCTS
Sakana AI research models, AI Scientist (autonomous research agent), evolutionary neural architecture search, Japanese-language model capabilities
HOW THEY GREW
Sakana AI published research on "AI Scientist" — an autonomous system that can conduct scientific research, generate hypotheses, run experiments, and write papers. The paper received significant media attention and positioned Sakana as a credible frontier research lab.
Being in Tokyo is deliberately differentiated: Japan has world-class engineering talent, a serious government AI agenda, and strong industrial partnerships.
THE HARD PART
Sakana AI is betting against the current consensus in AI research, which holds that scale (more parameters, more compute) is the primary driver of capability improvement. Building a successful research lab requires attracting top talent (competitive against DeepMind, Google, Anthropic, OpenAI), raising sufficient compute, and producing research that meaningfully advances the field.
The "small models" thesis is intellectually compelling but unproven at the frontier.
MONEY TRAIL
Seed
2023 · Led by Khosla Ventures, Lux Capital
$30M raised
Series A
2024 · Led by NEA, Sequoia Capital Japan
$30M raised
WHO BACKED THEM
Khosla Ventures, Lux Capital, New Enterprise Associates (NEA), Sequoia Capital Japan
Related Profiles
Head-to-Head
Compare Sakana AI vs another company.