The problem is that there are no open source models available (that can run on sub-$10k hardware) that match the capability of frontier models, or even "near-frontier" models like Minimax M3.
I tried it and found that I could run all of the <100B parameter models. They just didn't compare, so I returned the hardware.
Yeah, I don't know about that. I tested Qwen 3.6 27b, Qwen 3.5 122b, and Qwen 3.6 35b tonight and they gave me so much crap that I would have had to spend much more time cleaning up then had I just built it myself to start with.
Going back to opus 4.7. Very disappointed in open source options these days.
Edit: not that it matters a whole lot, but I'm on a MacBook pro M5 Max 128.
Hey @YouTube, now, when I see a new video from a creator that I don't subscribe to, I go to subscribe and it gives me an option of who I want to subscribe to. This is bad UI.