What it means
A language model reads everything you send, plus its own reply, within a fixed limit called the context window. It is measured in tokens, which are pieces of words. Many current models have windows of hundreds of thousands to about a million tokens, enough for long reports or whole contracts.
A bigger window is not free. Sending more text costs more and takes longer, and models can still miss details buried in very long inputs.
Why it matters for a business
The window decides whether a model can read a whole document at once or needs the text split and searched first. For large collections, retrieval is usually cheaper and more reliable than sending everything.
A business example
Things to watch
-
Longer inputs cost more and take longer.
-
Models can miss details in very long inputs.
-
Use retrieval when documents exceed what you want to send.