Rakuten Releases Japan’s Largest AI Model, Rakuten AI 3.0

Tokyo-based Rakuten has released Rakuten AI 3.0, described as Japan’s largest high-performance large language model, announced in a press release. The model was developed as part of the Generative AI Accelerator Challenge (GENIAC) project supported by Japan’s Ministry of Economy, Trade and Industry and the New Energy and Industrial Technology Development Organization.

Rakuten AI 3.0 is a Mixture of Experts model with approximately 700 billion parameters, optimized for Japanese language tasks. It is available free under the Apache 2.0 license via Rakuten’s official Hugging Face repository. The company reports that the model outperforms its previous versions in writing, code generation, and document analysis.

Part of the model’s training costs were supported by the GENIAC project, which provides computing resources for Japan’s generative AI initiatives. Rakuten stated that it evaluated the model across multiple Japanese benchmarks, including cultural knowledge, reasoning, and mathematics, showing improved accuracy compared to other leading Japanese-language models.

Rakuten AI 3.0 builds on the company’s prior models, Rakuten AI 7B and Rakuten AI 2.0, and continues its open-source approach aimed at advancing local AI development in Japan.