Share

Context Window

/ˈkɑːntekst ˈwɪndoʊ/

Definition

The maximum number of tokens (words) an LLM can process in a single interaction. Larger context windows enable deeper understanding of longer documents and more accurate responses.

Examples

  • GPT-4's context window of 128K tokens allows analyzing an entire book at once.
  • A small context window risks missing critical information.

Origin

Fundamental constraint of the Transformer architecture since 2017.

Critical metric determining LLM capability. Modern models increasingly support very large context windows.

TECHNeutral2017AILLM기술성능