PLaMo翻訳 update takes on frontier AI models at low cost

Preferred Networks built PLaMo翻訳 on the third generation of its domestic LLM PLaMo, giving the service a base that is updated from the prior version. The company says it performed translation-specialized training with several hundred times the compute of the previous generation, a scale of investment that would be difficult for smaller teams to match. That training appears to be the foundation for its accuracy claims against Claude Fable 5.1, GPT-6.1 Sol and GPT-6 Astra, though those comparisons rest on Preferred Networks' own evaluations, including its in-house benchmarks.

The update also pushes PLaMo翻訳 beyond a two-language tool: it now spans 53 languages, adding Chinese, Korean, Italian, Spanish, Thai and Russian among others. This broadens the service's potential use from Japanese-English documents to multilingual business settings, where a single provider covering many language pairs can simplify procurement. At the same time, the company pairs that expansion with a stated cost of about 14 yen per 100,000 characters of source text, which it says is under one-tenth of overseas frontier models and machine translation services. The two moves together suggest an attempt to compete not just on quality but on cost per unit of translation work.

The separately launched meeting translation feature adds a real-time layer for web conferences, with a bot joining Zoom, Google Meet and Microsoft Teams calls to show transcriptions and text translations per speaker, and with a glossary for proper nouns and technical terms. It covers 15 languages and can handle three-language meetings. How much this matters for business users hinges on whether the accuracy holds up outside Preferred Networks' own benchmarks and whether the stated cost remains stable at scale; those are the questions likely to shape adoption among procurement and localization teams evaluating the service.