I‘ve tried the very early Q4 mlx release on an M1 Max 32GB (LM Studio @ default settings), and have run into severe issues. For the coding tasks I gave it, it froze before it was done with reasoning. I guess I should limit context size. I do love what I‘m seeing though, the output reads very similar to R1, and I mostly agree with its conclusions. The Q8 version has to be way better even.