Context window
Also called: context length
The context window is a model's working memory — the maximum amount of text, measured in tokens, it can consider at once: your instructions, the conversation so far, whatever documents you've handed it.
The context window is a model’s working memory — the maximum amount of text, measured in tokens, it can consider at once: your instructions, the conversation so far, whatever documents you’ve handed it. Everything the model “knows” about your situation right now has to fit inside it. Anything outside might as well not exist.
Why it matters: This is why the brilliant collaborator turns amnesiac: a chat remembers a conversation, not a project, and as threads run long, early context falls out or gets compressed. Bigger windows help and don’t repeal the rule — attention thins across huge contexts, and fitting two hundred pages in is not the same as two hundred pages getting used. The durable fix is usually a workplace move, not a window upgrade: work that lives in files a coding agent re-reads every session beats work re-pasted into a chat that forgets.