Labsby Glyph
Japan AI Safety Institute (J-AISI)
Japan's AI safety institute, launched on 14 February 2024 within IPA to develop AI safety evaluation methods and standards.
Part of Information-technology Promotion Agency
Every lab
8 of 232 labs
Matsuo-Iwasawa Laboratory, University of Tokyo
University of Tokyo deep learning lab working on world models, robotics and LLMs, with many spin-off startups.
Part of The University of Tokyo
National Institute of Informatics (NII) and LLM-jp
Japan's national informatics institute, established in 2000, whose LLM R&D center runs the open LLM-jp project.
Preferred Networks
Japanese deep learning company that created Chainer and Optuna and develops the PLaMo Japanese language models and its own AI chips.
RIKEN Center for Advanced Intelligence Project (AIP)
RIKEN AI research center launched in 2016, covering mathematical foundations of ML, scientific AI and societal AI.
Part of RIKEN
Sakana AI
Nature-inspired methods: evolutionary model merging and The AI Scientist, from a Transformer co-author.
SB Intuitions
SoftBank subsidiary developing homegrown Japanese large language models under the Sarashina name.
Part of SoftBank Corp.
Sony AI
Sony Group's AI research subsidiary, known for the GT Sophy racing agent and the FHIBE fairness benchmark dataset.
Part of Sony Group
Questions
What counts as a lab here?
A research organization that builds or studies AI or data systems and publishes: labs inside large companies (Google DeepMind, Meta FAIR, Microsoft Research), startups whose business is models (OpenAI, Anthropic, Mistral AI, DeepSeek), nonprofits (Ai2, EleutherAI, METR), university labs (MIT CSAIL, Mila, Berkeley), government institutes (the UK AI Security Institute, the US CAISI) and data systems groups (CWI, DuckLabs, TUM).
Where do the facts come from?
Founding year, headquarters and handles come from Wikidata, Wikipedia or the lab's own pages. Landmark contributions link to the paper or the lab's own post (arXiv titles are checked against arXiv). Leaders come from leaders.fru.dev, which cites a source for each role; models from models.fru.dev; papers from papers.fru.dev. Every fact on a lab page links its source.
How is the flagship model chosen?
Where the lab is ranked on models.fru.dev, it is its best-ranked model there this week. Otherwise it is the newest model on the lab's own model page.
How often is it updated?
Every Monday a run pulls leaders, models, papers, blogs and funding from the other fru.dev sites and logs what moved. On the first Monday of each month it also re-checks every lab's website and landmark links. Nothing is overwritten silently: every change is kept on the Changes page.