#prompt-caching

Technique that stores the internal computation of reused prompt prefixes across LLM calls, reducing costs by up to 90% and latency by up to 85% in applications with repetitive context.