However output tokens are 5-10 times more expensive. So it ends up a lot more even on price
Even more than that in practice once you factor in prompt caching