Context window

The maximum amount of text a model can read at once.

An upper bound on tokens per request. Exceeding it truncates earlier content or errors, so long documents must be summarized or split.