Context Window

A context window is the maximum amount of text, measured in tokens, that a large language model can consider in a single request. It covers everything the model sees at once: the system prompt, the conversation history, any documents pasted or retrieved, and the answer the model is generating. When the total exceeds the limit, older or less relevant material must be dropped or summarized, which is why long chats begin to forget earlier details. Vendors compete on this number because a wider window lets a model read a whole codebase, contract, or research paper without splitting it into chunks. Anthropic's Claude models and Google's Gemini models are marketed heavily on long-context handling, and OpenAI's GPT family has widened its window across generations. A larger window is not free. Cost and latency usually scale with the tokens you send, and accuracy can sag for material buried in the middle of a very long input. Retrieval-augmented generation exists partly to avoid stuffing everything into the window, since it fetches only the passages that matter. Treat the context window as working memory rather than storage; anything the model must recall across sessions belongs in a file or database.