How to get started

You have used ChatGPT or Claude through their apps, and you want more: a different model, a lower bill, your own data staying yours, or just to understand what is actually going on. Here is the map, shortest useful version.

There are hundreds of models, and they are interchangeable in a way apps are not. Nearly every model — from every lab — speaks the same API shape. Code written for one talks to the others with a one-line change. That is why this site can compare them side by side, and why you are never locked in.

You have three ways to run one. Use the maker's own service, like the apps you know. Rent from a provider — companies that serve models by the token, where the same downloadable model often costs several times less from one host than another. Or download a model with open weights and run it on your own machine, where nothing leaves your computer.

Start by deciding what you actually need. For most people the honest answer is: a good general model, cheaply, with follow-up questions working. The models catalogue sorts by what has been measured; each page gives a plain verdict, live prices, and whether it would run on hardware you own. If a term stops you, it links to a two-minute explainer.

A sensible first hour: pick one well-rated mid-sized model from the catalogue. Make an account with one provider that serves it. Spend under a pound asking it real questions from your actual work. You now know more about the market than most people shipping AI products — and you will know what to look for in the next model.

Where next: The catalogue · When to consider running models locally