Model API
A web service that runs a model for you and charges by the token, so you never manage GPUs yourself.
Also: inference API, serverless inference
With a model API, a developer sends a prompt to a web address and receives the model's reply, paying per million input and output tokens. The provider handles GPUs, scaling and updates. Model labs offer APIs for their own models, while inference platforms such as Together AI, Fireworks AI, Baseten and the chipmaker clouds serve open-weight models, often through an interface compatible with OpenAI's, so switching is easy. Dedicated endpoints, where a customer reserves GPUs for one model, sit between this and renting GPUs outright.