NII

The National Institute of Informatics (NII) is a Japanese inter-university research institute under ROIS/MEXT, led by Director General Sadao Kurohashi (ACL Fellow 2025, Kyoto University professor). NII operates Japan's academic backbone (SINET) and the CiNii scholarly database.

The LLM-jp project (est. 2023) is Japan's largest open collaborative LLM effort: 2,600+ participants from 60+ universities and 200+ companies. Funded through GENIAC/NEDO with compute from Google Cloud Japan and SAKURA Internet. The flagship LLM-jp-3-172B (Dec 2024, 2.1T tokens, 50% Japanese/50% English+code) was the largest fully open Japanese model at release, beating GPT-4 on Japanese MT-Bench. LLM-jp-4 (Mar 2026) introduced MoE (32B/3.8B active, 128 experts, 65K context, 11.7T tokens) and "thinking" variants.

LLM-jp is unique for releasing the full stack: weights, training corpus (llm-jp-corpus, the largest Japanese web corpus), specialized tokenizer (Unigram byte-fallback), fine-tuning recipes, and safety benchmarks (AnswerCarefully). The project is organized into working groups covering corpus construction, safety, multimodal, and real-world robotics interaction.