Because of outstanding execution of OpenAI technical folks. An execution that has nothing to do with Altman. Similarly Mistral 7B model has much better performances than others. There is some smart engineering plus finding the magical parameters that produce great results. Moreover, they have a lot of training power. Unfortunately here the biggest competitor would be a company that lost its way a lot of time ago: Google. So OpenAI look magical (while it is using mostly research produced by Google).