PLaMo 3 Translate brings four translation models to local devices

Preferred Networks has released four PLaMo 3 Translate model sizes, ranging from 0.8 billion to 31 billion parameters, for translation across 53 languages. The November 5 announcement describes downloadable weights and a demonstration of the smallest model running on a phone in airplane mode.

Choosing a size and format

The company lists 0.8B, 2B, 8B and 31B checkpoints. Its technical write-up also lists GGUF and MLX variants along with several quantization formats. That range matters: a model can be small enough for an experiment on a phone while a larger sibling needs substantially more memory. PFN’s phone demonstration is an internal demo, not a released consumer app.

PFN says the models accept surrounding context and glossaries and can distinguish plain from polite Japanese. Its quality comparisons are the company’s own results, including proprietary evaluations. We have not benchmarked translation accuracy or tested it on our readers’ hardware.

The license is not a standard open-source license

The PLaMo Community License permits private and noncommercial use under its terms. The technical write-up says qualifying commercial users with revenue or income of one billion yen or less must register in advance; organizations above the threshold need a separate agreement. Read the model card and complete license for eligibility and current requirements.

Local execution can keep text on a device when the entire workflow is local, but privacy depends on how an application is built and configured. Our view: compact translation models are useful infrastructure for travel and local workflows, provided their quality, supported language directions and licensing are checked for the actual task.