Sakana AI
A Tokyo lab pursuing nature-inspired AI — merging models evolutionarily rather than training bigger ones.
Highlights
- Evolutionary model merging — combine multiple open-weight LLMs to produce better models without retraining
- AI Scientist — autonomous research agent that proposes hypotheses, runs experiments, writes papers
- Japanese-language foundation model focus alongside multilingual research
- Open-weight model releases including EvoLLM, EvoVLM, and Tinyswallow series
- Headquartered in Tokyo with backing from NTT and Japanese government AI initiatives
- Approximately $200M+ raised at $1B+ valuation across seed and Series A
- Investor base: NEA, Khosla Ventures, Lux Capital, Sony Group, NTT
External link — opens sakana.ai in a new tab. Sakana AI is a third-party product; we are not affiliated with it.
About Sakana AI
What it is
Sakana AI is a research lab founded by David Ha, formerly of Google Brain, and Llion Jones, a co-author of the original Transformer paper. Its research is deliberately not about scale: evolutionary model merging, which combines existing open-weight models into better ones without retraining; biologically grounded architectures; and the AI Scientist, an agent that proposes hypotheses, runs experiments and writes up results. It also builds Japanese-language foundation models.
Why it's different
Almost the entire frontier is a scaling argument — more compute, more data, larger models. Sakana is one of very few well-credentialled labs arguing the opposite, that the gains are in architecture and in recombining what already exists. Evolutionary merging is the concrete expression: producing a stronger model from existing open weights for a tiny fraction of training cost. Whether this competes with frontier scale is genuinely unresolved, and the AI Scientist work in particular has drawn substantive criticism about the quality of what it produces. Read the papers rather than the summaries.
How people use it
Its practical output for most people is research and open models rather than a product. The merging work is directly usable by anyone fine-tuning open-weight models, since it offers a cheaper route to a specialised model than training one. Japanese-language work makes it particularly relevant to teams building for that market, where the major Western models are noticeably weaker.
Written by the n3os team. We are not affiliated with Sakana AI.
This listing was written from public information, without Sakana AI’s involvement. If you own it and something here is wrong — or you would rather not be listed at all — email us and we will correct or remove it.
Get the ones worth knowing about
We write one of these for every tool worth the trouble. Get the new ones, plus what we have found genuinely useful lately.
Your address goes to Buttondown, who send the emails on our behalf. One click unsubscribes, and the list is never sold or shared.