What it is
Llama is Meta's family of open-weight models, which you download and run on your own hardware or through cloud hosts. The line includes Llama 2, the Llama 3 family (3.1, 3.2 and 3.3, from 1B to 405B parameters) and Llama 4, natively multimodal mixture-of-experts models. Meta's developer site provides downloads through Hugging Face and Kaggle, model cards, safety tools such as Llama Guard, and guides for fine-tuning, quantization and deployment.
The weights are free to use under Meta's Llama licenses, which carry conditions and are not OSI open source. Meta's newer models are released under the Muse name, including the open-weight Muse Glimmer, so Llama is no longer Meta's latest open model and is best suited to existing projects.
Best for
- Self-hosting a model you can fine-tune and run privately
- Building on a model family with wide tool support
- Comparing against newer open-weight models