PFN Releases Open-Weight Translation LLM PLaMo 3 Translate, Bringing World-Class Translation AI to Edge Devices
Available in four sizes ranging from 0.8B to 31B parameters; Translates locally on smartphones or internal servers, completely bypassing the cloud
Correction (11:45 11/05/2026): Under the PLaMo Community License, individuals and enterprises with an annual revenue of
less than1 billion Japanese yen or less in their most recent fiscal year can download and use the models free of charge.
TOKYO, November 5, 2026 – Preferred Networks, Inc. (PFN) today announced the open-weight release of PLaMo™ 3 Translate*, a suite of translation-specialized large language models (LLMs). Originally developed for PFN’s PLaMo Translate service, the open-weight models are now publicly available on PFN’s Hugging Face page.
Under the PLaMo Community License, individuals and enterprises with an annual revenue of 1 billion Japanese yen or less in their most recent fiscal year can download and use the models free of charge. This applies to both commercial and non-commercial applications, allowing users to run the models entirely on edge devices such as smartphones, laptops, and internal corporate servers without requiring an internet connection.
PLaMo 3 Translate provides users with the underlying model weights necessary to run the AI natively on their own hardware. It is available in four sizes: 0.8B (0.8 billion parameters), 2B, 8B, and 31B. This tiered approach allows developers to select the ideal model based on their specific requirements for translation quality, latency, processing volume, and target hardware.
By integrating PLaMo 3 Translate into device applications, users can achieve high-quality, high-speed translations across 53 languages including English and Japanese, without needing an internet connection.
PFN's release of PLaMo 3 Translate aims to expand recognition of its generative AI foundation model PLaMo, while accelerating open innovation through edge-device LLMs. PFN expects this release to drive new use cases across the tourism, enterprise business, disaster response sectors and more.
Key Features of PLaMo 3 Translate
1. High-quality output on small models
According to PFN’s comparative evaluations, even the smallest 0.8B version of PLaMo 3 Translate outperforms comparable translation-specific LLMs, such as Google’s TranslateGemma and Tencent’s Hy-MT2. Users can generate highly accurate translations locally without requiring massive computing infrastructure.
2. Scalable across hardware ecosystems
The availability of four distinct model sizes ensures that users can optimize for speed, quality, and hardware capacity. The smaller models can run swiftly on smartphones, laptops, and wearable devices completely offline, delivering instant translations. Use cases scale from personal mobile devices to on-premises enterprise servers and embedded corporate services.
3. Uncompromised privacy for confidential data
Because the model runs locally on user-managed servers or devices, neither the original text nor the translated output is ever sent to external cloud servers. This makes PLaMo 3 Translate ideal for translating highly sensitive materials, such as contracts, R&D documentation, and customer support logs, while maintaining strict compliance with corporate data privacy and security policies.
4. Advanced glossary control and contextual formatting
Users can customize translation outputs by feeding the model surrounding context and providing specific glossaries for proper nouns, technical jargon, and product names. Additionally, when translating into Japanese, the model offers precise control over the tone of the output, allowing users to specify either formal honorific forms or plain language forms.
Model Availability and Commercial Licensing
All four PLaMo 3 Translate models are available today on the AI community hub Hugging Face under the PLaMo Community License. Interested organizations can test the translation speed and quality on their own devices before fully integrating it into their workflows.
For model downloads and detailed licensing terms, please visit the respective Hugging Face repositories:
Commercial use, including internal corporate operations and integration into commercial products or services, is permitted free of charge for organizations generating less than 1 billion Japanese yen (approx. $6.5M USD) annually. Organizations exceeding this revenue threshold require a commercial license agreement with PFN. For commercial licensing inquiries, please contact PFN via the inquiry form.
For a deep dive into the performance evaluations and the development history of the models, read the PFN Tech Blog post: PLaMo 3 Translate: World-Class Translation with Open Weights—Here’s How We Built It
*PLaMo 3 Translate is a specialized translation model developed by PFN baked on PLaMo 3 NICT, a pre-trained model developed through joint research between PFN and Japan's National Institute of Information and Communications Technology (NICT).
