Inference

The moment a model is "run" to produce an answer, as opposed to training, which happens only once. Every time you press Enter in ChatGPT, that's inference.

Example: The cost of inference is why AI subscriptions have usage limits.