Context window

The context window is how much text a model can hold in mind at one time — the running span of prompt, documents and prior conversation it can actually reason over before older material falls out of view. It is a useful hinge in the chain because it turns an abstract capability into a concrete hardware bill: a longer window means the model must keep more intermediate state resident, so context length translates almost directly into memory footprint and bandwidth pressure during serving. Demand for long context is climbing as users push whole codebases, filings and case histories into a single prompt.

If that memory tightens, long-context features are the first to become uneconomic, because the cost of serving a request scales steeply as the window grows — pricing some applications out entirely even when the underlying model is capable. The exposure sits with labs leaning hardest on long context, such as Moonshot AI, whose products lean on the ability to reason across very large inputs. Here the model's ambition and the memory constraint are the same problem seen from two ends.