What is an open-source LLM?

An open-source LLM is one where the company publishes the trained model's actual parameters — the "weights" — so anyone can download it, run it on their own hardware, and modify it, instead of only reaching it through the company's app or a paid API.

Quick honesty check: most of these aren't fully "open-source" in the classic software sense. You get the weights, but usually not the training data or the exact code used to build them, which is why a lot of people in AI argue "open-weight" is the more accurate term. Worth knowing so you're not misled by the label.

Why people bother self-hosting a model

• Privacy — run it entirely on your own servers, so nothing you type ever gets sent to a third party.
• Cost control — at high volume, running your own hardware can beat paying per-API-call.
• Customization — you can fine-tune it on your own data to specialize it for your use case.
• Independence — you're not stuck with one company's pricing changes, rate limits, or the risk they discontinue the model.

The tradeoff: you need the hardware and know-how to run it, and it's usually a step or two behind the best closed models like GPT or Claude on raw capability. Meta's Llama and Mistral's models are the best-known examples right now, and both offer versions small enough to run on a decent laptop.