The other thing is most people don't really care about price per token or whatever but how much it will cost to execute (successfully) a task they want.
It doesn't matter if a model is e.g. 30% cheaper to use than another (token-wise) but I need to burn 2x more tokens to get the same acceptable result.