Glossary
Context Window
The maximum amount of text an LLM can consider in a single input.
Definition
The context window is the maximum number of tokens (roughly, words) an LLM can process in one request — including the prompt, any provided documents, and the generated response. Anything outside the window is invisible to the model for that request.
Context
By 2026, frontier models routinely support 200K–1M+ token context windows, which changes what is architecturally feasible: entire codebases, long document sets, or full meeting transcripts can be included in a single prompt. This has reduced the need for retrieval in some cases, though for cost, latency, and knowledge-freshness reasons RAG still dominates enterprise deployments.