I'm not sure that this comparison makes much sense, as in case of Mistral you will have to rent the GPU 24/7 to be able to use it, but at the same time you won't use 100% of it all the time. In case of OpenAI API you pay only for what you use.
And let's not forget that even though mistal is very performant and promising model, it's still less accurate than gpt 3 and hallucinates more. (and I'm not even talking about gpt4).