IT Home On October 8, the Japanese company Preferred Networks (PFN) released its PLaMo 3 Translate 31B model, which specializes in Japanese and also handles translations in other languages, locally at yesterday. The modelwas entirely developed by Japan itself, and Japanese accounts for approximately 30% of the training data.
The model is based on the PLaMo 3 base model. It supports a number of languages expanding from 2 in the previous generation (Japanese and English) to 53, and introduces an online meeting translation mode that covers 15 languages.
PFN claims that its PLaMo 3 Translate 31B model has anaverage score superior to OpenAI’s GPT-6.1 Sol and GPT-6 Astra in four translation benchmark tests. Additionally, due to its efficient handling of Japanese, the cost of this model is only 14 yen per 100,000 characters of input (IT Home note: approximately 0.59 yuan at current exchange rate), which is significantly lower than competing models.