
Independent company · Tokyo, Japan
rinna
A Japanese generative AI company known for releasing pretrained Japanese GPT, BERT and speech models, including the Youri and Nekomata LLMs.
Tokyo, Japan
- Founded
- 2020
- Type
- Independent company
- Based in
- Tokyo, Japan
- Key papers
- 1
- Featured projects
- 4
About
rinna was established in 2020 when the team behind Microsoft's Japanese Rinna chatbot, launched in 2015, separated from Microsoft to operate independently.
Its research group releases base models for Japanese text, audio and images, including GPT, BERT, HuBERT, CLIP and Stable Diffusion variants, on Hugging Face and GitHub. Its early Japanese GPT-2 models became common baselines for Japanese NLP.
The Youri series continues pretraining of Llama 2 on about 40 billion tokens of mixed Japanese and English data, and the Nekomata series builds on Alibaba's Qwen models with a tokenizer that is more efficient for Japanese.
Mission and ambitionsrinna aims to give Japanese developers open pretrained models for their language.
Known for
- 01
Japanese GPT-2 models
- 02
Youri models
- 03
Nekomata models
- 04
Japanese pretrained models
Milestones
- 2020
Company established
Rinna team spun out of Microsoft Japan.
- 2021
Japanese GPT-2
Released small, medium and xsmall Japanese GPT-2 models.
- 2023
Youri
Released Llama 2 based Japanese-English models.
- 2023
Nekomata
Released Qwen-based models with a Japanese-efficient tokenizer.
Open source
Most downloaded model.
Japanese CLIP.
Japanese RoBERTa.
- japanese-gpt2-mediummodel
Japanese GPT-2.
Research papers
Notable releases
Openness
Weights on Hugging Face and code on GitHub.
Contact and community
Contact via its website or Hugging Face page.
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 Something wrong or out of date?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.



