Why am I paying more for the narration of a text response than the production of the text response, when the model for the latter is usually orders of magnitude larger?
The price difference is so massive I can't point to batching as the reason either, it just seems like the pricing got anchored with massively higher margins than text generation.