Japan Selects 7 Domestic LLMs for Government Deployment

🔑 Enhanced Key Takeaways

• •Fujitsu's Takane LLM was successfully piloted in 2025 with a central government agency, automating public comment classification and summarization with over 80% accuracy in linking opinions to draft law clauses[1].
• •Japan's first National Basic Plan for AI, approved February 2026, emphasizes accelerating AI in government operations, strengthening domestic models, and expanding the Japan AI Safety Institute[2][3].
• •NTT's tsuzumi 2 is a lightweight LLM deployable on a single GPU, achieving top Japanese language performance in business domains while prioritizing data sovereignty for on-premises use[6].

🛠️ Technical Deep Dive

• •tsuzumi 2: Lightweight model optimized for Japanese, excels in financial/medical/public sectors, supports RAG and fine-tuning, matches frontier models on single GPU with world-top results among similar-sized models[6].
• •Takane: Demonstrated 80%+ accuracy in identifying relevant clauses between draft laws and public opinions; enables automation of opinion classification by support/opposition and summarization[1].

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline