Context Window

The context window is one of the most practically important specifications of any large language model. Everything the model can 'see' at once - your question, all prior conversation turns, any documents you've provided, and the system prompt - must fit within this window. Text outside the window is simply invisible to the model, as if it doesn't exist.

Context windows are measured in tokens. Early GPT-3 had a context window of 4,096 tokens (roughly 3,000 words). Modern models have expanded dramatically: GPT-4 Turbo offers 128,000 tokens, Claude offers up to 200,000 tokens, and some research models are exploring million-token contexts. Larger context windows allow models to analyze entire books, long conversations, and extensive codebases without losing context.

The context window determines what the model can coherently respond to. In a long conversation, once the conversation history exceeds the context window, the oldest messages are dropped. This is why AI assistants sometimes seem to 'forget' what was discussed at the beginning of a very long conversation. Applications that need to maintain long-term context use techniques like conversation summarization (compressing old context into summaries) or retrieval-augmented generation (storing information externally and retrieving relevant parts as needed).

Context window length also affects model performance. Models typically attend to the beginning and end of the context most reliably, with information in the middle of very long contexts sometimes receiving less attention - a phenomenon researchers call the 'lost in the middle' problem. Structuring important information at the beginning or end of your prompt is a practical tip for working with long contexts.

For professionals using AI tools, context window size determines what documents you can analyze in one go. A model with a 200,000-token context window can read and reason about a full research report, entire contracts, or extensive codebases in a single session. This capability is transforming how professionals use AI copilots for document analysis, code review, and research tasks.