I would like to override the global settings for models that support prompt caching.

Currently there are no way to disable prompt caching for some models like Gemini, that could be very expensive as it is billed by hour.