Oh boy.
GDM is making (or has been backed into a corner into making) the bet that high throughput, low latency, low capability models are the path forward.
That probably works for vibe coded apps by non-practitioners.
I suspect that practitioners/professionals will wait longer for better results.