Context window
The amount of text a language model can consider in a single request — measured in tokens, roughly three-quarters of a word each.
A context window is the maximum amount of text a language model can hold in its "working memory" for one exchange. Measured in tokens (roughly 3/4 of a word each), it caps how much you can send in and how much the model can send back combined.
If you paste a document longer than the context window, the model literally cannot see the whole thing at once — either the tool truncates the input silently, or it refuses to run, or (in a well-designed product) it warns you.
Practical thresholds worth memorizing: - 8K tokens ≈ a short article - 32K tokens ≈ a chapter - 128K tokens ≈ a novella - 200K tokens ≈ 300 pages / a full technical document / most single filings
The context window is not the model's "memory across conversations" — that's a separate feature usually called memory or projects. The window is per-conversation, per-request.
Bigger windows are not always better. They cost more compute, so providers charge more for them. And past a certain point, models get worse at attending to the specific parts of the input you actually care about ("lost in the middle"). Modern frontier models handle 200K tokens well, but if your task fits in 30K, using 30K will be faster and often more accurate.