I agree.

I might be wrong about this, but obviously Google would like to provide inferencing at the lowest cost to themselves, so perhaps their slow ‘pro’ releases and rapid ‘flash’ releases is an attempt to guide people to use more profitable models?