No, why would they. You are supposed to maintain that cache.

What I really want to know is about caching the large prefixes for prompts. Do they let you manage this somehow? What about llama and deepseek?