Comment by CaveTech

Your question is worded kind of confusingly, but all caching is handled on the inference layer, and by all major providers. In short, caching should work as long as you are sending requests to the same model and provider.