# Semantic caching

Reusing a prior answer or result for a semantically similar request to reduce repeated model computation.

Distinct from exact-match caching — it catches requests that are worded differently but mean the same thing. One of the handful of AI-specific cost-optimization practices the FinOps Foundation names explicitly, alongside model tiering and routing.

**Related research:**
- [Token Economics: The Atomic Unit of AI Value](https://measuredvalue.ai/research/token-economics-atomic-unit-of-ai-value/)