Rakuten Launches Open Japanese LLM 3.0

🔑 Enhanced Key Takeaways

• •Rakuten AI 3.0 is an approximately 700 billion parameter Mixture-of-Experts (MoE) model, Rakuten's largest to date[1][2].
• •Developed as part of Japan's GENIAC project by METI and NEDO, with partial training cost support for computing resources[1].
• •Achieves top scores in Japanese language benchmarks and up to 90% cost reduction compared to third-party frontier models in Rakuten services[1].
• •Preceded by Rakuten AI 2.0, an earlier MoE model released in February 2025 that was 8x more powerful than its predecessor[2].

🛠️ Technical Deep Dive

• •Model size: Approximately 700 billion parameters[1][2].
• •Architecture: Mixture-of-Experts (MoE) optimized for Japanese language and culture[1][2].
• •Training data: Leverages open-source community models and Rakuten’s high-quality bilingual data[1].
• •Performance: Top scores in Japanese language benchmarks; up to 90% cost reduction vs. third-party models in internal trials[1].

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline