Inference is the model doing the work right now. Training already happened; inference is the model spending that training the moment you hit send. It can feel like the AI learns from every answer; in normal use it runs the model without updating its brain. A run can include reasoning (extra thinking steps before the answer) and streaming (the answer appearing word by word).
A trained chef cooking tonight’s dinner captures it. The chef spent years learning, but tonight’s meal depends on the order you place, the allergy notes you share, and the ingredients in front of them. Training built the skill; the cooking happens now.
Why you care
Inference is where cost, speed, and quality meet. Every run uses compute: the more context you send, the more it reads; the longer the answer, the more it writes. Input matters for the same reason a great chef handed stale ingredients still makes a poor dinner: a strong model with weak context gives weak output. You don’t need the strongest model to sort simple items with clear rules, but you probably do for strategy, ambiguous writing, or a messy codebase.