What is an 'open' large language model versus a closed one?

An open model means the actual code and trained 'weights' — the files that make the AI work — are publicly available for anyone to download, modify, and run on their own computer. A closed model keeps those files secret. You can only access it through a website or paid API. DeepSeek-v3 is a good recent example of an open model. You can download it right now and run it on a powerful enough machine. ChatGPT, by contrast, is closed. You can't download it. You can't peek inside. The difference matters more than most people realize. With an open model, a hospital could run patient data through it without ever sending sensitive information to some company's server. A developer in Brazil could fine-tune it to speak better Portuguese. No permission needed. The trade-off is that open models usually require technical skill and decent hardware. Running DeepSeek-v3 on a single GPU is possible, as recent research shows, but it's not as simple as opening a web browser. Closed models handle all that complexity for you. You just type and get results. I think of it like cooking at home versus eating at a restaurant. Open models give you the recipe and ingredients. Closed models give you a great meal but you never see the kitchen. Neither is universally better. It depends entirely on whether you need control and privacy, or convenience and ease.