Token Cache Hit

A billing concept for large model APIs. When the input prefix of a request is exactly identical to that of a previous request, the server reuses the already cached computation results, and that portion of tokens is billed at a lower cache-hit price.