How does Anthropic prompt caching primarily reduce cost and latency on repeated requests?
Select an answer to reveal the explanation.
Short Explanation
Put the big stable stuff up front — system prompts, tools, docs — mark it cacheable, and later calls reuse that prefix cheaply.
Full Explanation
Prompt caching lets you designate large, reusable prompt prefixes (system instructions, tool definitions, RAG context, long documents). After a cache write, subsequent requests that share the same prefix read from cache at reduced token pricing and with lower time-to-first-token. It is not permanent multi-week conversation storage, client GPU weights, or skipping tokenization.