https://blog.google/technology/ai/lamda/

It's supposed to be ahead of OpenAI's current GPT. Seems only logical to me since it was Google that invented and open-sourced the Transformer model and they likely have much more and much higher quality data than OpenAI for training.