Table of Contents
What Makes Repeated Claude Requests with Large Prompts Cheaper?
Prompt caching cuts costs and latency for repeated large system prompts in Claude API—ideal for production scale over unrelated features like citations or thinking.
Question
You’re making many requests with the same large system prompt. What feature would make your requests faster and cheaper?
A. PDF processing
B. Citations
C. Extended thinking
D. Prompt caching
Answer
D. Prompt caching
Explanation
Prompt caching in the Claude API optimizes repeated requests sharing the same large system prompt by storing its internal KV cache on Anthropic’s servers, slashing latency by up to 90% and costs by up to 90% on cache hits since only new conversation content gets processed from scratch.
When your identical system prompt prefix (e.g., role instructions, examples) matches a prior request within the cache window (typically minutes to hours), subsequent calls reuse the cached computation, making it ideal for high-volume apps like customer support bots or code assistants with static guidelines—far superior to PDF processing (A, document parsing), citations (B, sourcing), or extended thinking (C, internal reasoning traces).