PFN opens the weights of PLaMo 3 Translate: a 0.8B model runs 53 languages offline
Synopsis
Preferred Networks released PLaMo 3 Translate, a translation-specialized LLM suite originally built for its own PLaMo Translate service, as open weights on Hugging Face in four sizes (0.8B, 2B, 8B, 31B), runnable entirely offline on smartphones, laptops or internal corporate servers across 53 languages including English and Japanese. Under the PLaMo Community License, individuals and enterprises with annual revenue of 1 billion yen or less in their most recent fiscal year can download and use the models free for commercial and non-commercial purposes; larger organizations need a commercial license. PFN says its comparative evaluations put even the smallest 0.8B version ahead of comparable translation-specific LLMs such as Google's TranslateGemma and Tencent's Hy-MT2.
Interpretation
The weights of a translation-specialized LLM can now be downloaded and run on local hardware instead of only being reached through a cloud service. These models previously served as the internal basis of PFN's own PLaMo Translate service and were offered only as a service. All four sizes are published on PFN's Hugging Face page under the PLaMo Community License, and organizations can test translation speed and quality on their own devices before integrating them into workflows.
The models come in four parameter tiers so developers can pick by translation quality, latency, processing volume and target hardware. The tiers are 0.8B, 2B, 8B and 31B, with smaller models aimed at smartphones, laptops and wearables and larger ones at on-premises enterprise servers and embedded services. PFN's comparative evaluations state that even the smallest 0.8B version outperforms comparable translation-specific LLMs such as Google's TranslateGemma and Tencent's Hy-MT2.
Local inference keeps both the source text and the translated output off external cloud servers. Cloud-based translation previously required sending the text to be translated to an external server. PFN names contracts, R&D documentation and customer support logs as suitable sensitive material and says this deployment meets corporate data privacy and security compliance requirements.
Perspective
The report covers the four model sizes, the 53-language coverage, the community license terms and the self-reported comparative evaluation; performance conclusions apply only to the comparison setting PFN describes.
The release gives no specific metric values or benchmark sets for PFN's comparative evaluation, so the claim that the 0.8B version beats TranslateGemma and Hy-MT2 rests on the vendor's own account; the license is divided by whether revenue in the most recent fiscal year exceeds 1 billion yen.