Llama is Meta's family of large language models whose weights can be downloaded, so developers can run and adapt them on their own hardware.
In plain words
The first LLaMA, in February 2023, came in sizes from 7 to 65 billion parameters and was shared with researchers; it showed strong models could be trained on public data alone. Llama 2 added chat-tuned versions, and Llama 3 grew to a 405-billion-parameter model. Llama 4, in April 2025, switched to a mixture-of-experts design and reads images as well as text. The weights come under Meta's own community licence, which allows commercial use with conditions.
A quick example
Imagine a company that cannot send customer emails to an outside AI service. It downloads a Llama model, fine-tunes it on its own support replies, and runs it on its own servers.
Where you'll meet it
- Self-hosted chatbots and assistants.
- A base for community fine-tunes and research.
Related
Sources
- PaperLLaMA: Open and Efficient Foundation Language Models, Touvron et al. (arXiv) · read 29 Sept 2026
- PaperLlama 2: Open Foundation and Fine-Tuned Chat Models, Touvron et al. (arXiv) · read 29 Sept 2026
- PaperThe Llama 3 Herd of Models, Llama Team, Meta (arXiv) · read 29 Sept 2026
- DocsLlama-4-Scout-17B-16E-Instruct model card, Meta (Hugging Face) · read 29 Sept 2026
- Docsllama-models, Meta (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.