Having used both Google's and OpenAI's models, the kind of issue they have are different. Google's models are superior or at least on par in knowledge. It's the instruction following and understanding where OpenAI is significantly better. I don't think pretraining data is the reason of this.