Japan's PFN launches PLaMo 3 Translate, beating OpenAI models in translation tests

Japanese AI company Preferred Networks (PFN) has launched a translation model called PLaMo 3 Translate 31B. It focuses on Japanese while also handling many other languages. The model was fully developed in Japan, with Japanese making up about 30% of its training data.

Built on the PLaMo 3 base model, it now supports 53 languages, up from just Japanese and English before. It also adds an online meeting translation mode covering 15 languages, which is handy for cross-language meetings.

In tests, the model scored higher on average than OpenAI's GPT-6.1 Sol and GPT-6 Astra across four translation benchmarks. It is also cheap, charging just 14 yen for every 100,000 input characters, far less than its rivals.