Providers

When you use ChatGPT you are talking to one company's model on one company's computers. Aninference provider rents you the same thing directly: you send text to an address on the internet, their machines run the model, and the reply comes back. That address is the API, and the tools you already use — an editor extension, a chat app, your own script — talk to it for you.

It matters because the same model is sold by several of them at different prices, different speeds, and different rules about whether your prompts are stored or trained on. The rows below are those differences.