Rakuten AI 3.0: Pioneering the Future of Language Models in Japan
Rakuten Group has unveiled its latest innovation in artificial intelligence: Rakuten AI 3.0, a Japanese large language model (LLM) that exemplifies advancements in the field of generative AI. This development is part of the Generative AI Accelerator Challenge (GENIAC) project, which is endorsed by Japan’s Ministry of Economy, Trade and Industry (METI) and the New Energy and Industrial Technology Development Organization (NEDO).
Technical Architecture: Mixture of Experts
At the heart of Rakuten AI 3.0 is its innovative Mixture of Experts (MoE) architecture. This model features approximately 700 billion parameters, making it the largest LLM developed by Rakuten to date. The MoE structure allows the model to activate around 40 billion parameters for each token processed. While the remaining parameters stay inactive, this design enhances efficiency and substantially reduces computational costs. During trials, Rakuten AI 3.0 demonstrated a remarkable 90% cost reduction compared to other leading third-party AI models when integrated into the Rakuten Ecosystem.
Seamless Integration Across Services
Rakuten AI 3.0 will be accessible through the Rakuten AI Gateway, which serves as the company’s internal platform for a suite of generative AI APIs. This integration promises to enrich various Rakuten services, enhancing user experience and operational efficiency. Furthermore, an open-weight version of the model is slated for release in Spring 2026, which could further stimulate innovation in the generative AI landscape.
Foundation and Training
The development of Rakuten AI 3.0 incorporated insights from the open-source community while leveraging Rakuten’s extensive bilingual datasets. Its training was conducted on an in-house, multi-node GPU cluster tailored specifically to meet the requirements of this ambitious project. Notably, the model’s training benefits from the support provided by the GENIAC initiative, which is designed to bolster generative AI development in Japan.
Tailored Efficiency for Japanese Language Processing
What truly sets Rakuten AI 3.0 apart is its specialized focus on the Japanese language and culture. Fine-tuned for conversational abilities and instruction-following tasks, the model underwent evaluations using the Japanese MT-Bench. This benchmark assesses various capabilities—such as maintaining persona consistency and generating engaging content—allowing Rakuten to gauge its performance relative to previous models, including Rakuten AI 2.0 and other global benchmarks.
Security and Data Protection
Security and data privacy remain paramount in the deployment of Rakuten AI 3.0. The model operates within Rakuten’s secure cloud environment, which emphasizes stringent data control. All proprietary data used for fine-tuning is kept within the company’s internal framework, ensuring that user data and corporate information stay safe from external breaches, a crucial concern in today’s digital landscape.
Commitment to Innovation and Collaboration
Rakuten’s dedication to advancing artificial intelligence aligns with broader industry objectives. According to Ting Cai, Chief AI & Data Officer of Rakuten Group, the company aims to roll out cost-effective, high-quality models that can benefit a diverse range of users, from small businesses to large enterprises. This philosophy is echoed by Watanabe Takuya, Director of the AI Industry Strategy Office at the METI, who expressed enthusiasm about the potential of Rakuten’s advancements to foster a robust AI ecosystem in Japan.
Future Implications and Developments
Rakuten’s commitment to continuous improvement in the field of AI and LLMs suggests that we are only witnessing the beginning of a transformative journey. By developing LLMs in-house, Rakuten is not only securing its technological advancements but is also building a wealth of knowledge and expertise that will enable it to create increasingly sophisticated AI-driven services tailored for the unique requirements of the Japanese market.
In summary, Rakuten AI 3.0 is a testament to the company’s relentless pursuit of innovation in artificial intelligence, uniquely positioned to enhance digital interactions and drive economic growth within Japan’s burgeoning AI industry.