LLM-jp (NII) Open Models: Model List and Licenses
About LLM-jp (NII)
According to its organization card, LLM-jp brings together more than 1,000 researchers and engineers in natural language processing and computer systems from universities and companies, under the auspices of the National Institute of Informatics (NII). Its goals are to build open-source LLMs proficient in Japanese together, share research information and release models, tools and technical materials. The latest LLM-jp-4 is developed by NII’s Research and Development Center for Large Language Models.
What It Releases
- LLM-jp-4 (2026): 8B (March), the 32B-A3B MoE (March) and 33B (August), each as base, instruct and thinking versions. According to the model card, the base models use pre-training and mid-training only, and the post-trained models are aligned with SFT and DPO, without reinforcement learning.
- LLM-jp-4-VL 9B (August 2026): a model that also reads images, combining llm-jp-4-8b-thinking with SigLIP 2. Unlike the beta, it was trained only on data with no license or terms-of-use issues for commercial use.
- Speech: the kaburi-tts speech synthesis model (July 2026) and llm-jp-moshi for spoken dialogue.
- It also releases many intermediate and research checkpoints, such as sparse-model studies (the organization has 268 models).
Licenses
The LLM-jp-4 line is Apache-2.0, broadly usable including commercially. Research models based on Llama (such as Llama-Mimi) use the original Llama license. Read each model’s license text before use. This summary is not legal advice.
Running Them Locally
- The thinking versions have official GGUF builds. LLM-jp itself publishes GGUF files for the llm-jp-4 8B, 32B-A3B and 33B thinking models.
- On a consumer GPU, the 8B or 32B-A3B are the candidates. The 32B-A3B is an MoE with about 3B active. Usage examples are collected in the official cookbook (llm-jp-4-cookbook on GitHub).
Sources: the llm-jp organization card; model cards of llm-jp/llm-jp-4-8b-thinking and llm-jp/llm-jp-4-vl-9b; the model list of the llm-jp organization on Hugging Face (all as of 2026-09-26).
Our Coverage and Data
Local Model Watch has not published an article on LLM-jp (NII)’s models yet. The model list below is collected by code from Hugging Face every day. Part of our publisher index; model families link to their family pages.
At a Glance
| Item | Value |
|---|---|
| Hugging Face organization | llm-jp |
| Models on Hugging Face | 268 |
| Articles on its own models | 0 |
| Articles on derived models | 0 |
Recent Models on Hugging Face
The publisher lists 268 models in total on Hugging Face. Below are up to 15 of the most recent generative models (text, image, video, audio and so on), the scope of this site, per organization. Classifiers, feature extractors, research components and repositories without a model card are omitted, and the publisher’s own quantized builds and helper variants are folded into the row of the original model. Parameter counts are left blank for repositories that pack weights into integer types, where Hugging Face’s count is not the real parameter count.
→ Scroll horizontally to see all columns
| Released | Model | Task | Parameters | License | Official variants |
|---|---|---|---|---|---|
| 2026-08-15 | llm-jp/llm-jp-4-vl-9b | vision-language | 9.1B | apache-2.0 |
— |
| 2026-08-14 | llm-jp/llm-jp-4-33b-thinking | text generation | 33.2B | apache-2.0 |
gguf |
| 2026-08-14 | llm-jp/llm-jp-4-33b-base | text generation | 33.2B | apache-2.0 |
— |
| 2026-07-21 | llm-jp/kaburi-tts | audio | — | apache-2.0 |
— |
| 2026-03-30 | llm-jp/layoutlmv3-japanese-preview | — | 111M | apache-2.0 |
— |
| 2026-03-30 | llm-jp/llm-jp-4-8b-instruct | text generation | 8.6B | apache-2.0 |
— |
| 2026-03-16 | llm-jp/llm-jp-4-32b-a3b-thinking | text generation | 32.1B | apache-2.0 |
gguf |
| 2026-03-16 | llm-jp/llm-jp-4-32b-a3b-base | text generation | 32.1B | apache-2.0 |
— |
| 2026-03-16 | llm-jp/llm-jp-4-8b-thinking | text generation | 8.6B | apache-2.0 |
gguf |
| 2026-03-15 | llm-jp/llm-jp-4-8b-base | text generation | 8.6B | apache-2.0 |
— |
| 2026-02-24 | llm-jp/llm-jp-3.1-8x13b-32K | text generation | 73.2B | apache-2.0 |
— |
| 2026-02-24 | llm-jp/llm-jp-3.1-8x13b-32K-instruct4 | text generation | 73.2B | apache-2.0 |
— |
| 2026-01-07 | llm-jp/llm-jp-moshi-v1 | — | 7.7B | apache-2.0 |
— |
| 2025-09-18 | llm-jp/Llama-Mimi-8B | audio | 8.1B | llama3.1 |
— |
| 2025-09-18 | llm-jp/Llama-Mimi-1.3B | audio | 1.3B | llama3.2 |
— |
From the Hugging Face API. Last changed 2026-09-27 (JST). Collected by code.
Last updated 2026-09-27 (JST). The explanation at the top of this page was written with the help of AI from the primary sources it cites. The tables and lists under “Our Coverage and Data" are assembled by code from our article log and the Hugging Face API.