The only thing I can figure is that Google's consumer-facing free-to-use models are set to "reason as fast as possible with only the most basic plausibility checks." I have to correct them like this every single time.
Similarly, you can run local LLMs on your Google Pixel smartphone, but if you want them to be hardware accelerated using the chip that's included for that very purpose on your phone... it has to be Google's models via Google's apps. So you can't say "alright, a fast, local model which sounds great, but I want one that is trained to always run a search before answering, in a harness with good search tools for answering the kinds of practical questions I often ask", you're actually not free to put that together in a practicable way.
So there is indeed some competition here. But it's not really fair competition.