Learn
What is prompt caching?
Last reviewed July 16, 2026
Prompt caching is a provider feature that stores the processed form of a prompt's prefix (system instructions, attached documents, conversation history) and reuses it when later requests start with the same prefix. Cached input tokens bill at a fraction of the normal rate and skip reprocessing, so long-context, many-turn workloads get cheaper and often faster.
How caching works
Models process prompts left to right, so providers can snapshot the computation after a stable prefix. A cache hit requires the prefix to match exactly: put stable content first (instructions, reference documents) and volatile content last (the newest user message). Caches expire in minutes on most providers, so rapid consecutive requests benefit most.
What it does to cost
Cache-read rates are typically a large discount against normal input rates, and some providers bill a premium to write the cache. The economics favor long shared prefixes queried repeatedly: agents re-reading the same instructions every step, or a document interrogated across many turns. Rates differ per provider and appear as separate cache-read rows on pricing tables.
Model pricing pages in idapt list cache read rates beside input and output prices where a provider offers them, and per-message usage shows what a reply actually cost.
Frequently asked
See it in practice
The fastest way to understand it is to use it. Start free, no credit card.
Try idapt free