September 2026
Intermediate
206 pages
5h 13m
English
Prompt compression reduces the number of tokens you send. Prompt caching reduces the price of the tokens you send. If your system prompt, tool definitions, or common context prefixes are the same across many requests, you’re paying full price for the same tokens thousands of times a day. Caching lets you pay once to write those tokens into a cache and then read them back at a ~90% discount on every subsequent request.
The impact is dramatic. In a production system I worked on, enabling Anthropic prompt caching on a 2,000-token system prompt across 200,000 daily requests reduced input token costs by approximately 85% for those tokens. The engineering effort was minimal — restructure ...
Read now
Unlock full access