Prompt caching
Saving the processed form of a long, repeated prompt so later requests that start the same way are cheaper and faster.
Also: context caching
Many applications send the same long opening text, such as a system prompt, a manual or a codebase, with every request. Prompt caching keeps the KV cache for that shared prefix on the provider's servers for a few minutes or longer, so later requests skip reprocessing it. Anthropic, OpenAI, Google and DeepSeek introduced forms of it in 2024, typically charging much less for cached tokens. It works only when the repeated text comes first and matches exactly.