Rakuten Releases 700B Parameter Open-Source AI Model

Subscribe to Our Newsletter

Success! Now Check Your Email

To complete Subscribe, click the confirmation link in your inbox. If it doesn’t arrive within 3 minutes, check your spam folder.

Ok, Thanks
Rakuten Releases 700B Parameter Open-Source AI Model
Source: Rakuten

Rakuten Releases 700B Parameter Open-Source AI Model

Rakuten launches Rakuten AI 3.0, a 700-billion parameter MoE model developed under Japan's GENIAC project, available via Apache 2.0.

Philip Lee profile image
by Philip Lee

Tokyo, Japan - Rakuten Group, Inc. has released Rakuten AI 3.0, an open-source artificial intelligence model with approximately 700 billion parameters.

The model was developed through the Generative AI Accelerator Challenge (GENIAC), a program funded by Japan's Ministry of Economy, Trade and Industry (METI) and the New Energy and Industrial Technology Development Organization (NEDO) to support domestic generative AI development.

The model uses a Mixture of Experts (MoE) architecture and is available for download under the Apache 2.0 license through Rakuten's Hugging Face repository.

According to Rakuten, it is the largest model developed domestically in Japan, following the company’s earlier models, Rakuten AI 7B (approximately 7 billion parameters) and Rakuten AI 2.0 (approximately 47 billion parameters).

Rakuten stated that the model was trained using open-source models and the company’s bilingual data, with optimization for Japanese cultural knowledge, historical context, and linguistic characteristics.

Potential applications include automated text writing, code generation, and document analysis, according to the company.

In performance tests using Japanese-language benchmarks, Rakuten AI 3.0 scored 76.9 on JamC-QA (cultural and historical knowledge) and 72.1 on M-IFEval (instruction-following), exceeding OpenAI’s gpt-4o in both metrics, according to the company.

On the MMLU-ProX (graduate-level reasoning) and MATH-100 (competition-level mathematics) benchmarks, Rakuten AI 3.0 scored 71.7 and 86.9, respectively.

The GPT-OSS-Swallow-120B-RL-v0.1 model scored 77.5 and 96.0 on those benchmarks.

The GENIAC project partially funded the development of Rakuten AI 3.0 and provided computational resources after Rakuten was selected for the program's third phase in July 2025.

Ting Cai, Chief AI & Data Officer at Rakuten Group, stated that the company is developing large language models and plans to work with METI to establish an open AI development community in Japan.

The release is part of Rakuten’s corporate strategy, which the company refers to as "AI-nization," aimed at integrating AI technologies across its operations.

Philip Lee profile image
by Philip Lee

Subscribe to The Pickool

Success! Now Check Your Email

To complete Subscribe, click the confirmation link in your inbox. If it doesn’t arrive within 3 minutes, check your spam folder.

Ok, Thanks

Read More