The context window is the maximum quantity of text, measured in tokens, that a model can process in one interaction, including both the input supplied and the output generated. Anything outside it is unavailable to the model.
Windows have grown substantially, but performance often degrades before the stated limit, particularly for information positioned in the middle of very long inputs.
Alternative Names:
Context Length, Input Window
Why it Matters?
Litigation matters routinely exceed any context window. A single deposition can consume a large share of one, and a matter file is orders of magnitude larger. This is the technical reason retrieval architecture exists: rather than attempting to fit a case into the window, the system selects the relevant passages for each question. Claims that a tool reads an entire case file at once should be examined closely.
Frequently Confused with
Related terms
Frequently asked questions
Can a large context window replace retrieval?
How much text fits in a context window?
