
Independent company · Tokyo, Japan
Stockmark
Stockmark is a Tokyo company that has released large Japanese-language models, including the 100-billion-parameter Stockmark-100b, under permissive licenses.
Tokyo, Japan
- Founded
- 2016
- Type
- Independent company
- Based in
- Tokyo, Japan
- Featured projects
- 7
About
Stockmark is a Japanese company focused on business information and language models. Its co-founder and CTO is Kosuke Arima, and its researchers publish models on Hugging Face under the stockmark organization.
Stockmark-100b is a 100-billion-parameter model pretrained from scratch on about 910 billion tokens of Japanese and English text and released under the MIT license. The company later released Stockmark-2-100B-Instruct and a multimodal Stockmark-2-VL-100B, both developed under Japan's GENIAC program.
The models target Japanese business use, with a stated focus on reducing hallucination in domain questions.
Mission and ambitionsStockmark aims to provide Japanese-first language models that companies can run and adapt, with open weights.
Known for
- 01
Stockmark-100b Japanese model
- 02
Stockmark-2 100B series
- 03
MIT-licensed Japanese LLMs
Milestones
- 2024
Stockmark-100b
A 100B Japanese and English model trained from scratch is published under the MIT license.
- 2025
Stockmark-2-100B-Instruct
A second-generation 100B model is developed under GENIAC.
- 2025
Stockmark-2-VL-100B
A vision-language follow-up is released.
- 2025
Availability on NVIDIA NIM
Stockmark-2-100B-Instruct is offered through NVIDIA's model catalogue.
Open source
Support page for a book on NLP with BERT.
Japanese NER dataset.
Unofficial MIND recommender implementations.
Japanese GPT-NeoX.
- stockmark-13bmodel
13B Japanese model.
- OmniDocBench-JASyndataset
Japanese doc benchmark.
- ner-wikipedia-datasetdataset
NER dataset.
Notable releases
Openness
Weights on Hugging Face.
People
- Kosuke ArimaCo-founder and CTO
- Takahiro OmiResearcher and model author
Listed from the lab's own pages and public profiles.
Contact and community
Contact via its website.
Sources
Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.



