Meta Llama

Llama is Meta’s family of open-weight large language models — models whose trained weights you can download, inspect, fine-tune, and run on your own hardware. Released in successive generations at multiple sizes, from small models that fit on a laptop to large mixture-of-experts flagships, Llama comes with a permissive (though not fully open-source) community license that allows commercial use for almost everyone. It’s arguably the model family that made serious local and self-hosted AI mainstream.

Llama matters because open weights change your relationship with the model. With a hosted API you rent intelligence; with Llama you can own the deployment — run it in your own VPC for data-sensitive workloads, fine-tune it on your domain, quantize it to fit cheaper hardware, and never worry about a vendor deprecating your model or raising prices. The Llama ecosystem is also enormous: most local-inference tools, fine-tuning frameworks, and research papers support it first, and countless derivative models on Hugging Face are Llama fine-tunes.

In practice, you’ll rarely train Llama yourself. You’ll pull weights from llama.com or Hugging Face, run them locally through Ollama or LM Studio for prototyping, and serve them in production with an inference engine like vLLM behind an OpenAI-compatible endpoint. Alternatively, hosted providers such as Groq, Together, and AWS Bedrock serve Llama per-token — giving you open-model economics without managing GPUs.