Rakuten Develops New Large Language Model; ‘Rakuten AI 3.0’ Is One of Japan’s Biggest Llms to Date

The Yomiuri Shimbun

2:00 JST, December 19, 2025

Rakuten Group Inc. has developed its own Japanese large language model (LLM), a foundational model for generative artificial intelligence, the company announced on Thursday. It plans to gradually introduce the model, one of the biggest yet developed in Japan, into its online shopping services and other platforms.

According to the announcement, Rakuten’s LLM, “Rakuten AI 3.0,” achieved approximately 700 billion parameters, a metric indicating training scale. It is said to possess a deeper understanding of Japanese-specific nuances, culture and customs than other large language models. Operational costs were reduced by up to 90% compared to other models of similar scale. The company plans to make the model available externally around next spring.

Domestically, firms such as NTT Inc. and a SoftBank subsidiary are also developing their own LLMs. SoftBank’s “Sarashina” model achieved approximately 460 billion parameters in fiscal 2024.

The development of Rakuten’s LLM was supported by “Geniac,” a project of the Economy, Trade and Industry Ministry and other entities to aid domestic AI development. Rakuten received subsidies to help cover the costs of using the graphic processing units required to create a large language model.