PLaMo 3 Translate: PFN's open translators for 53 languages
Preferred Networks released PLaMo 3 Translate in 0.8B to 31B sizes for 53 languages, with GGUF files from 0.57 GB and a license that limits commercial use to 1 billion yen of revenue.
Preferred Networks (PFN), a Japanese AI company, published the weights of PLaMo 3 Translate on Hugging Face and described them in a blog post dated 5 November 2026. There are four sizes, 0.8B, 2B, 8B and 31B, and they cover 53 languages under the PLaMo Community License.
What is new
The models start from PLaMo 3 NICT, a pretrained model that PFN built jointly with Japan’s National Institute of Information and Communications Technology, and PFN trained them further on tens of billions of tokens of translation data. Much of that data was written with Gemma 4: PFN synthesized translations, fused candidates in a tournament, back-translated text and refined the results through iterative self-distillation. Two rounds of preference optimization followed, and the models up to 8B were also distilled from the 31B.
The target language is required, and the source language is detected if you leave it out. The card supports surrounding context, a glossary and a choice between plain and polite Japanese. Input and output together are limited to 32,768 tokens.
PFN’s own charts compare the models with TranslateGemma and Hy-MT2. They report every PLaMo size ahead on Japanese-English and multilingual tests, even the 0.8B model against models more than 30 times larger. Part of that evidence is PFTB, an internal Japanese-English set whose references were written with LLMs.
Running it
Every size comes as BF16 weights, as FP8, W4A16, W4AFP8 and NVFP4 builds for vLLM, as GGUF files for llama.cpp and as 4-bit MLX models for Apple silicon. The 31B model is 63.5 GB in BF16 and 18.7 GB in NVFP4, the 8B model is 17.1 GB and 5.3 GB at Q4_K_M, and the 0.8B model has a 0.57 GB GGUF. PFN’s tables show most formats close to BF16, with 2-bit GGUF losing the most.
For production PFN recommends the vLLM Completions API with a hand-built prompt in its own format. The Transformers chat template needs custom code and is meant for trials. PFN showed the 0.8B model translating on a smartphone in an internal demo and has released no app.
License
The PLaMo Community License is free for personal and academic use. Commercial use is free after registration while annual revenue or sales, including affiliates, stay at or below 1 billion yen, and above that you need a separate agreement. Products must carry the notice “Built with PLaMo”, derived models must have “PLaMo” in their name, and models trained on the outputs count as derived models.
Why it matters
Open translation models now come from Google, Tencent, Bilibili and PFN, and they differ in languages, controls and licenses. PLaMo is the one tuned for Japanese, with the smallest builds, and it is the only one of the four with a revenue limit.
The quality claims are PFN’s own, and we found no independent comparison. Our PLaMo 3 Translate vs Hy-MT2 page sets out the trade-offs, and the translation models list shows the whole field.
Sources
- PLaMo 3 Translate 31B model card (Hugging Face) huggingface.co
- PLaMo 3 Translate: World-Class Translation with Open Weights (PFN blog, 5 November 2026) preferred.jp
- PLaMo Community License Agreement (text in the model repository) huggingface.co
- PLaMo 3 Translate 0.8B model card (Hugging Face) huggingface.co
Models in this story
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| PLaMo 3 Translate Preferred Networks | 890M – 31.8B | 32K | PLaMo Community (Conditional) | 0.614 GB Laptop or phone | |
| Hy-MT2 Tencent | 2.04B – 30.1B 3B active | 256K | Apache 2.0 (Permissive) | 1.41 GB Laptop or phone | |
| TranslateGemma Google DeepMind | 4.3B – 27.4B | 2K | Gemma Terms (Conditional) | 2.97 GB Laptop or phone | |
| Index-Translate Bilibili | 2.27B – 36B 3B active | 256K | Apache 2.0 (Permissive) | 1.44 GB Laptop or phone |