The context window
The context window is the total amount of text (measured in tokens) the model can "see" at once - your entire conversation, any documents you've attached, and its own response, all counted together. Once you go over the limit, the oldest parts get cut or the request fails. Bigger context window = the model can consider more information at once, but it doesn't mean better reasoning.
Context windows range from tens of thousands to over a million tokens depending on the model. Everything inside it competes for the model's attention - very long contexts can cause the model to "lose track" of details buried in the middle (sometimes called the "lost in the middle" effect). This is why good prompt engineering often means putting critical instructions at the start or end, not buried in a wall of text.
Paste a long document into Claude (5,000+ words) along with one specific instruction buried in the middle of it, and one at the very end. Ask a question that requires both. Notice which one it follows more reliably - a live demonstration of context window behavior.
You bury one instruction in the middle of a 5,000-word document and put another right at the end. Which is the model more likely to follow reliably?