The maximum amount of tokens (input + output budget) a model can attend to in one request. Larger windows help, but unmanaged context still hurts quality, latency, and cost.
The context window is your working memory budget for a single call: system instructions, retrieved documents, tool results, and conversation history all compete for the same space.
Practical implications
Very long prompts dilute attention; irrelevant text can derail reasoning (“context poisoning”). Patterns like summarization, RAG, caching, and offloading large tool payloads keep the window focused.
See also
Model Context Protocol for tool and resource access without stuffing everything inline.