Llama (Meta)

Llama is the open language model family from Meta AI. Since 2023 it has shaped the open-source LLM ecosystem like no other – hundreds of derivatives and fine-tunes build on Llama weights.

Current versions

The latest open generation is Llama 4 (April 2025) with the Scout and Maverick variants: natively multimodal, mixture-of-experts architecture, 17 billion active parameters each. Meta’s newest frontier model (April 2026), by contrast, launched as a closed, API-only release for the first time.

Why it is so popular

Llama brought open weights into the mainstream. A large community, broad tool support, and many specialized derivatives lower the barrier to entry – anyone evaluating open models rarely gets past Llama.

Self-hosting entry point

Llama models run locally via tools like Ollama or vLLM and on every major cloud provider. That makes them the typical starting point for self-hosted AI.

License specifics

Llama uses the Meta Community License – not a genuine open-source license under the OSI definition. Among other things, it requires a special license above 700 million monthly active users and imposes naming rules for derivatives. Uncritical for most companies, but worth checking.

Last updated: June 2026