Updated Aug 20, 2026

Model

The trained artifact itself — the file of learned numbers that turns an input into an output.

Share

What it means

In everyday AI usage a model is the concrete, saved result of training: a large file of learned parameters, plus the architecture describing how to run them. It is the thing you download, host, call through an API, or pay per use.

The distinction worth holding is between the model and the product. ChatGPT is a product; the models behind it are separate artifacts that get swapped as new ones ship. A vendor can change the model under a product without renaming the product, which is exactly why capability and pricing can shift without any visible announcement.

A model is also frozen. It does not learn from your conversations, and it knows nothing that postdates its training — hence the knowledge cutoff, and hence retrieval as the standard way to give it current information.

Why it matters

Confusing the model with the product leads to bad decisions: benchmark results attach to models, but you use products. Knowing which model a product is running — and whether that can change beneath you — is a real procurement question, not a technicality.

In practice

When comparing tools, ask which model each one calls and whether you can pin it. An application that silently upgrades models is convenient until an upgrade changes behavior you'd built around.

Related terms