Context Window: Definition and Examples

The context window is one of the fundamental concepts for understanding how large language models (LLMs) work. It represents the maximum limit of tokens — words, subwords, or characters — that a model can "see" and process simultaneously during an interaction. This window includes everything: the system prompt, conversation history, input documents, and the response being generated.

Practically, think of the context window as the model's working memory. Everything inside this window is accessible to the model for formulating its response. Anything beyond this limit is simply invisible. For example, if you provide a 200,000-token document to a model with a 128,000-token window, a significant portion of the document will be truncated or ignored.

Context window sizes have evolved considerably. Early GPT-3 models had about 4,096 tokens, while recent models like Claude offer windows up to 200,000 tokens or more. This progression has transformed possible use cases: analyzing long documents, extended conversations, processing entire codebases.

In prompt engineering, managing the context window effectively is a key skill. It's not enough to have a large window — you need to use it wisely. Placing the most important information at the beginning and end of the prompt, summarizing previous exchanges, and structuring data concisely are techniques that maximize response quality while respecting size constraints.