Self deployment vs Roboflow inference prices

What is the difference if I want to deploy all models trained in roboflow to my server vs accessing the model through the API from roboflow directly?

Do I have to pay less if it is self deployed?

Anybody can answer to my question?