← Concept Index

Reasoning & inference

Inference — running the model

Also called: using vs training, prompt to completion, serving

DEFINITION

Inference is what happens every time you use a model: your prompt goes in, the trained weights are run over it, and a completion comes out. No learning occurs — the weights are fixed.

WHY IT MATTERS

Separating inference from training clears up most confusion: your prompts don't teach the model, each request is (by default) independent, and the cost and speed you experience are inference costs, paid per use.

COMMONLY CONFUSED WITH

Training. Training builds the model once, at great expense; inference is the cheap, repeated act of using it — and it's the only part most people ever touch.