Feb 19, 2026 13:02:00

NVIDIA releases 'Nemotron-Nano-9B-v2-Japanese,' a small AI model specialized for Japanese, achieving the highest performance under 10B and ready for commercial use

NVIDIA has released ' Nemotron-Nano-9B-v2-Japanese, ' a small-scale language model (SLM) specialized for Japanese. Nemotron-Nano-9B-v2-Japanese achieves the highest performance of any Japanese model with fewer than 10 billion parameters, and has been released as an open model for commercial use.

NVIDIA Nemotron 2 Nano 9B Japanese: Cutting-edge small-scale language model supporting Japan's sovereign AI

https://huggingface.co/blog/nvidia/nemotron-nano-9b-v2-japanese-ja

At the time of writing, most AI services run models on the cloud, but companies that handle confidential data must run models on their own in-house servers. This creates a demand for compact, high-performance models that can run on in-house servers, but there remain few options for compact models with high Japanese language processing capabilities.

Nemotron-Nano-9B-v2-Japanese is a model customized based on Nemotron-Nano-9B-v2 using the Japanese dataset Nemotron-Personas-Japan and other data, and is positioned as an SLM with advanced Japanese language processing capabilities.

Nejumi Leaderboard, which aggregates the results of Japanese language performance measurements of AI models, ranks models with less than 10 billion parameters as follows: Nemotron-Nano-9B-v2-Japanese achieved the top score, beating out models such as Qwen3-8B and Gemma-3-12b-it.

Below is a graph summarizing the performance of each category of 'Qwen3-8B', 'Nemotron-Nano-9B-v2', and 'Nemotron-Nano-9B-v2-Japanese'. Nemotron-Nano-9B-v2-Japanese outperforms the base model and competing models in knowledge tests and coding tests.

Nemotron-Nano-9B-v2-Japanese is also characterized by its high-speed processing, and in tests performed by NVIDIA, it recorded speeds up to six times faster than the Qwen3-8B.

The model data for Nemotron-Nano-9B-v2-Japanese is available at the following link. The license is the commercially available ' NVIDIA Nemotron Open Model License .'

nvidia/NVIDIA-Nemotron-Nano-9B-v2-Japanese · Hugging Face
https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2-Japanese

The Japanese dataset, Nemotron-Personas-Japan, is also available at the following link. The license is Creative Commons Attribution 4.0 International .

nvidia/Nemotron-Personas-Japan · Datasets at Hugging Face
https://huggingface.co/datasets/nvidia/Nemotron-Personas-Japan