Reference · Glossary

Context window

The maximum amount of text (measured in **tokens**) a model can read and remember **in one conversation turn**.

When to use

Planning long documents, multi-file code help, or chat history — if input is too long, the model drops or forgets middle parts.

When not to

Don't confuse with long-term memory products; the context window is the working desk, not a filing cabinet.

Example

A 128K-token window fits a short book chapter plus your questions. A 500-page PDF may need chunking or RAG instead.