That would be very bad product design. My understanding is that the model itself is similar to GPT4o in architecture but trained and used differently. So the 5x relative increase in output token cost likely already accounts for hidden tokens and additional compute.