to be fair, the new models are far more efficient. Open weight models are eating their lunch though, they have to keep giving more to slow the exodus.
Its great for sure, if you have the hw to run it! If not you're left spinning up GPU instance (which I do!) but for the average bear they just something that works!
for now, but both model makers and hardware developers will work to get you to use models which are good enough to run on their devices without any special GPU requirements