Rakuten Releases Japan's Largest High-Performance AI Rakuten AI3.0; Community Finds It's Based on DeepSeek V3

Rakuten Group recently released Rakuten AI3.0, claiming it as Japan's largest high-performance AI model and open-sourcing it for free under the Apache 2.0 license. The model uses a Mixture of Experts (MoE) architecture with a total of 671 billion parameters, activating 37 billion parameters per inference, a 128K context window, and is optimized for Japanese—it outperforms GPT-4o in multiple Japanese benchmark tests.

This model is a result of the GENIAC project jointly promoted by Japan's Ministry of Economy, Trade and Industry (METI) and the New Energy and Industrial Technology Development Organization (NEDO), receiving partial government computing power funding. However, the community found by checking the model configuration file on HuggingFace that its underlying architecture is actually DeepSeek V3, with parameters exactly matching DeepSeek V3, indicating it was fine-tuned with Japanese data based on DeepSeek V3.