I’m considering DeepSeek for a workload that sends the same long system prompt with different user messages. Before changing the request design, I’d like to know whether its usage information makes it possible to see when repeated prompt content is being reused.

For anyone using the API in production: which usage fields or dashboard details do you check to compare cached and uncached requests? I’m trying to connect DeepSeek prompt cache hits with actual changes in latency and billed tokens, rather than assuming repeated text is automatically saving anything.