PLaMo-3-8B

PLaMo-3-8B is the 8-billion-parameter base model of Preferred Networks' PLaMo 3 family, a sovereign Japanese LLM line co-developed with NICT and released in 2025. It uses a Gemma-based architecture with sliding window attention, trained on 1 trillion tokens split across English, Japanese, code, and multilingual data, with a 4,096-token context.

The family spans 2B, 8B, and 31B, all released as open base weights on the pfnet Hugging Face organization. Community work has shown the 8B model runs well with standard tooling and supports chat-style templates with function-calling extensions.

It ships under the PLaMo community license rather than a standard OSI license, so commercial users should read the license terms on the model card. Preferred Networks also operates the separate PLaMo 3.0 Prime enterprise platform with reasoning models and on-premise deployment for businesses.