Every AI model boils down to a large set of numbers produced by training, stored in files. Whether a model is open or closed comes down to who gets those files.
With a Closed model: A model you can only use through its maker's apps or API: the files that make it work stay private. Most ChatGPT, Claude and Gemini models are closed., the maker keeps them to itself. You use the model through the maker's app or website, or through its API: How developers plug a model into their own apps and pay per use. A model that's "in the API" can be used by programmers, even if it isn't in a chat app yet. if you're a developer. Your messages travel over the internet to computers you don't control. The main models behind ChatGPT, Claude and Gemini work this way.
It's like eating at a restaurant: you can order the dish, but you can't take the recipe home, and the menu can change without warning. In return, you get the maker's latest work with nothing to install.
Open-weight models
With an Open weights: The model's files are public, so anyone can download it, run it on their own computer and build on it, usually for free. trckr labels these "Open". model, the maker publishes the files for anyone to download, usually for free and often on Hugging Face: The website where labs and researchers publish models anyone can download. Think of it as an app store for AI models.. You can run the model on your own computer, change it with extra training or build it into a product. trckr labels these models “Free to download”.
In restaurant terms, open weights are a ready meal you can take home, reheat and season as you like, but you still don't get the recipe. People often call these models Open source: Strictly, the model's files, code and training recipe are all public and free to reuse. Often used loosely to mean "free to download"., but strictly that means the recipe is shared too: the training code and details of the data. Many “open” models share only the finished files.
Free to download doesn't mean free to use however you like: every open model comes with a licence, the rules for using it. Some licences, such as Apache 2.0 or MIT, let you do almost anything, including selling a product built on the model. Others ban commercial use, add conditions for very large companies or rule out certain uses. Check the licence before you build a business on one.
Why run a model yourself
Privacy is the main reason. When a model runs on your own machine, your words and files never leave it. That matters for doctors, lawyers and businesses with confidential data, and for anyone who'd simply rather keep things private.
Cost and control matter too. Once you have the hardware there's no charge per question, nobody can change or withdraw the model you rely on, and it works without an internet connection. You can also train it further on your own material, so it learns your field or your style.
What you need
The computer you need depends mostly on the model's size, measured in Parameters: The numbers a model learns during training: a rough measure of its size. Bigger usually means more capable, but slower and pricier to run. 7B means 7 billion.. Small models with a few billion parameters can run on a recent phone or laptop. Medium-sized ones need a desktop computer with a good graphics card, the kind gamers use. The largest, with hundreds of billions, need servers full of expensive chips, so most people use them through an online service even though they're free to download.
Shrunk-down copies, called Quantisation: Shrinking a model by storing its numbers less precisely, so it fits on smaller computers. A little less accurate, much lighter. FP8, GGUF, AWQ and MLX are common formats. versions, help a model fit on a smaller machine at the cost of a little quality. Free desktop apps can download and run one for you in a few clicks, so you don't need to be a programmer to try.