OpenAI is a weird animal. Their research seems generally less impactful than rest of the top tier labs out there. But they do have great people on the team. Is it because their engineering infrastructure is subpar?
As far as I can remember, most important research from OpenAI is InfoGAN. Others feel somewhat below outstanding. Their open source record on the other hand is pretty solid, gym/universe and various implementations and testing of existing models does live up to the openness in their name.