Meta's open-weight LLMs — the backbone of the self-hosted, open-source agent stack.
Category: Foundation Models · Type: Open-weight model family · Website: https://www.llama.com
Overview
Llama is Meta's family of openly licensed model weights (a permissive community license rather than strict OSI open-source) that anyone can download, fine-tune, and self-host. It is the most popular foundation for open-source agents and for teams that need on-prem or cost-controlled inference. The current generation is Llama 4 (Scout and Maverick, mixture-of-experts and multimodal).
Key Features
- Open weights for self-hosting and fine-tuning
- Range of sizes from small/edge to large frontier-class
- Huge ecosystem of fine-tunes and tooling
- Runs on Groq, OpenRouter, and local runtimes
Role in the Agentic AI Stack
The default open 'brain' for self-hosted agents; frequently served through fast-inference providers like Groq and routed via OpenRouter.
Part of Agentic AI Landscape → Foundation Models