Explainer
Context window
Everything a model can take into account at once, measured in tokens.
When a chatbot writes a reply, the only information about your conversation it can use is what sits in its context window. That includes hidden instructions from the app, the conversation so far, any files you added and the reply it is writing. Anything about your conversation that falls outside the window does not exist for the model at that moment. The window has a maximum size, and when a long chat reaches it, apps drop or summarize older parts. Everything else the model draws on comes from training and stays fixed; the window is short-term and changes with every message. A large window helps, though models do not use every part of a long context equally well.
Going deeper
Liu et al. (2023) found models used relevant information best near the start or end of a long input and worst in the middle (arxiv.org), and Anthropic's documentation calls the general decline in accuracy as context grows "context rot" (platform.claude.com).