Cache hit (prompt caching)

Also: prompt caching · cached input

A cache hit is when a model reuses a previously processed prompt prefix instead of recomputing it, billed at the discounted cached-input price. For agents with a stable system prompt or shared context, cache-hit rate is frequently the single biggest lever on cost per run — often larger than the choice of model.

← All terms · Definitions maintained by Mintok. Methodology.