Lotte has unveiled "Japan's largest high-performance AI model," with its specifications indicating that it is based on the DeepSeek V3 architecture

Svmuu News: Rakuten Group has released the open-source model Rakuten AI 3.0, describing it as "Japan's largest high-performance AI model." The model employs a MoE architecture with a total of 671 billion parameters, activates 37 billion parameters per inference, and features a 128K context window. Optimized for Japanese language scenarios, it outperforms GPT-4o in multiple Japanese benchmark tests. The model is one of the outcomes of the GENIAC project promoted by Japan’s Ministry of Economy, Trade and Industry (METI) and the New Energy and Industrial Technology Development Organization (NEDO), and received partial computing power support. Rakuten did not disclose the source of the base model, stating only that it was built upon open-source community contributions. The community discovered in the HuggingFace model file that its `config.json` contains "deepseek_v3" and related architecture fields, and that the parameter size and context configuration match those of DeepSeek V3, indicating that the model may be based on DeepSeek V3 with fine-tuning for Japanese.