The Responses API is the new stateful API. Given this new option, when would you still use cache prompting if the Responses API can “save” the initial message/prompt?
For instance, let’s say we have a long text with many small, independent queries (perhaps one per paragraph of the long text). Using prompt caching, I would need to send the long text each time at the beginning of the prompt to access the cache, followed by the actual query for the paragraph. With the new Responses API, I would send the long text only once and then use the response.id to make queries for the individual paragraphs (in this example). It seems to me that this new stateful API is actually a better implementation of caching, offering more control. Perhaps there is an even better way of solving this example.
However, I am unclear if the initial message (when the response was created) is considered cached in terms of pricing. Perhaps this is where OpenAI is providing options, but from a technical perspective, I don’t see why one would use cache prompting when the new Responses API is available.