PANews reported on March 17 that Rakuten Group announced the launch of its Japanese-specialized large-scale model, Rakuten AI 3.0. The company claims that this model outperforms models like gpt-4o in several Japanese language benchmarks, including those related to Japanese culture and history, and instructional compliance. It was developed based on open-source community models combined with Rakuten's own bilingual data.
However, users on X discovered that the config.json file of the Hugging Face project page contained content such as "model_type": "deepseek_v3". Furthermore, when answering related questions, the model's stance clearly leaned towards China rather than Japan. This sparked discussions among users about whether the model was developed based on DeepSeek and how much of its technology was independently developed in Japan.


