Yes. I kick myself that I did not get the 500gb model when getting was good. It was a massive price tag, but the hardware is NICE. And that MLX architecture really works. I run the gpt-oss 100gb model and it's right quick.

In my opinion - not yet. The local models are getting better all the time, but still aren't quite reliable enough. Also, hardware prices are absurd right now. Once the hardware market settles down and there's a little more progress on local models, maybe in a year or two, I expect things will be a bit different.

I'm running a Strix Halo with 128GB of RAM. It's fun to play with, but the real frontier models are just way better.