The Japanese artificial intelligence market has long been searching for its own direction. The large language models that are on everyone's lips today are predominantly Western or American developments. The Japanese language often remains a secondary priority for them: these models understand it, but with noticeable limitations. It is within this context that the emergence of Rakuten AI 3.0 should be understood.
What Happened
On March 17, 2026, Rakuten Group announced the release of its new language model – Rakuten AI 3.0. Its developers call it Japan's largest high-performance AI model. The model was created as part of the government program GENIAC (Generative AI Accelerator Challenge) – an initiative jointly promoted by Japan's Ministry of Economy, Trade and Industry and the state-owned organization NEDO, which focuses on the development of energy and industrial technologies.
The model was first publicly announced back in December 2025. Now, a refined, additionally trained version has been released, ready for real-world application.
Why This Is a Government Initiative, Not Just a Corporate One
GENIAC is not just about funding startups. It is a deliberate attempt by Japan to create its own generative AI infrastructure: to avoid complete dependence on foreign platforms and to have models that work well with the Japanese language, Japanese business culture, and the Japanese regulatory context.
Simply put: the government set the task, allocated the resources, and major players – in this case, Rakuten – took on the development. This is a fairly typical model of technological development for Japan, where industrial policy and business work in tandem.
In this sense, Rakuten is a logical participant. The company manages one of the largest ecosystems in the country, encompassing e-commerce, banking, telecom, and content. It has both the motivation to integrate AI into its own services and the data and technical infrastructure to do so.
Who Is This Model For?
Rakuten AI 3.0 is positioned not as a consumer product, but as a tool for companies and developers creating their own AI applications. In short, it's not something you launch in a browser to chat with a chatbot. It is a foundation upon which others can build their own solutions.
This approach makes sense: the Japanese corporate software market is huge, and demand for reliable tools with strong Japanese language support is consistently high. Banks, insurance companies, media, and retail – all are potentially interested in integrating a language model into their processes without having to explain the nuances of Japanese business correspondence to an American system.
What Remains Behind the Scenes
The official announcement does not disclose technical details – neither the exact size of the model, the data it was trained on, nor comparative benchmarks against other systems. The title of “Japan's largest high-performance model” remains a declaration for now, one that is difficult to verify without independent testing.
This doesn't necessarily mean there is nothing behind the words. But it is a signal that the model's true assessment has yet to be made – not by a press office, but by those who will use it in practice.
A separate question is how well such public-private initiatives can compete with global developments in the long term. The Japanese market is substantial in size, but also quite specific. Whether this specificity will be enough for Rakuten AI 3.0 to carve out a stable niche, only time will tell.
For now, the fact remains: Japan now has its own large language model, developed domestically, with state support, and tailored to the Japanese business environment. For an industry accustomed to looking to Silicon Valley, this is at least a notable step in a different direction.
