Llama models ship open weights you can run free
Meta releases Llama models with open weights, so local runs cost nothing beyond your own hardware.
What is free right now
Meta keeps releasing Llama models with open weights. Once the weights are public, the model itself costs nothing to run: you download it and serve it on your own laptop, your own server or a free cloud notebook. There is no per-token charge and no subscription, which makes open-weight releases the most durable kind of free access in AI.
How to use it free
The simplest path is a quantized build through a local runner such as Ollama or LM Studio, which runs comfortably on consumer hardware for the smaller sizes. If your machine is too weak, free-tier notebooks on Google Colab can host the mid-size models for experiments. Always download from the official Llama site or a mirrored distribution page so you get the real weights.
What to watch out for
The license is open but not unrestricted: commercial use at very large scale has extra terms, and some regions are excluded. Bigger checkpoints need serious memory, so match the model size to your hardware. Open weights do not mean open training data, and the free license can change between versions, so read the current one before shipping anything commercial.
⚠️ Each item states the free facts and how to use them free. No credit amounts are promised; free rules can change at any time, so treat the official pages as the source of truth.