04 · How Models Run

Inference

Running a trained model

Running a trained model to produce an output. Every prompt-response interaction uses inference; for heavily used products it can become a major recurring cost, while the balance between training and serving costs varies by system.

Concrete example

Each reply a chatbot gives you is one act of inference, with a real per-query cost to whoever runs it.

Why it matters

Where most of the ongoing cost of an AI product actually lives.