What the Context Window Is

Context window

Part of: The Context Window

Imagine answering a question while staring at a desk that can only hold a few sheets of paper at once. Anything you want to consider has to fit on that desk. Push a new sheet on and an old one slides off the edge. That desk is the context window : the model's entire field of view for a single request. What it is The context window is the maximum number of tokens a model can consider in one request. And here's the part people miss: that budget covers both what you send in and what the model writes out. Input and output share the same desk. A token is just a chunk of text, roughly 4 characters of English, about ¾ of a word. So a window of, say, 8,000 tokens is room for roughly 6,000 words of combined prompt and answer. Modern models advertise windows from a few thousand tokens up into the millions, but the rule is identical at every size: there is a hard ceiling, and everything must fit under it. How it works Think of the window as the model's working memory , not its long-term knowledge. Its long-term knowledge is baked into frozen weights during training. The context window is the scratch space it gets right now , for this request: Two things follow immediately. First, the bigger y

Challenge: Context Packer