Prompt Caching

Cache large, repeated prompt prefixes (system prompts, long documents, tool definitions) to cut latency by up to 85% and reduce input token costs by 90% on cache hits.