Context window
The amount of text an AI model can consider at one time — its short-term memory. Everything you send and everything it replies has to fit inside this window, measured in tokens. A larger context window lets a model work with longer documents and conversations.
A language model does not remember earlier conversations the way a person does. Each time it responds, it can only take into account the text placed in front of it right now: your latest question, the earlier back-and-forth, and any documents you have pasted in. The size of that working space is the context window, and it is measured in tokens.
This is why a model can lose track of something you mentioned much earlier in a long chat, or refuse a document that is simply too big. Once the conversation grows past the window, the oldest material falls out of view. Early models could hold only a few hundred words; modern ones can hold the equivalent of whole books, which has opened up tasks like summarising a lengthy contract in one go.
The context window also explains why techniques like retrieval-augmented generation exist. Rather than trying to cram an entire knowledge base into the window, a system fetches just the few relevant passages and slots them in alongside your question, making the most of the limited space.