What does it actually mean when people say an AI model is 'open source'?

It means the creators have shared the model's underlying code and 'weights' publicly, so anyone can download, study, change, and run it. Think of it like a recipe. A closed-source AI is like a restaurant that sells you a cake but won't share the recipe. An open-source model like Meta's Llama is like a chef who publishes the full recipe online. You can bake the cake yourself, tweak the ingredients, and even sell your modified version. DeepSeek-v3.2 is a recent example of a powerful model pushing this open frontier. The real magic for beginners isn't just the free access. It's that you can run these models on your own computer, without sending your data to a company's server. This matters for privacy and for tinkering. I've found that the term gets fuzzy, though. Some 'open' models still have restrictive licenses. For instance, you might be able to use it for research but not for a commercial app. Always check the license file on the model's Hugging Face page. A useful tip: don't assume 'open source' means 'easy to use.' Setting up these models on your own machine often requires a bit of technical comfort with the command line. The community around a model is just as important as the code itself, because that's where you'll find the guides to actually get it running.