OpenRouter's explanation of implicit caching makes clear that its zero-data-retention classification can still permit temporary prompt-derived caches. These caches keep intermediate model calculations so repeated input can be processed with less computation. The company treats such temporary caching as compatible with its classification when there is no other persistent storage.
That is a platform policy interpretation, not a promise that no prompt-derived information ever remains after a request. The article acknowledges that cached representations need protection and says customers can inspect endpoint caching information and adjust routing if their requirements exclude it.
Terms in this piece · Glossary
prompt caching — Reusing the model's processed form of a repeated prompt prefix so subsequent calls skip re-reading it, cutting cost and latency substantially.
KV cache — The memory a model keeps about text it has already read, so generating each new token doesn't require reprocessing the whole conversation.
