Literally every single one of these papers was developed with an internal model that not even most OpenAI employees have access to.