I had it running on my 128gb strix halo - it ran around 40 tokens per second I think but I found it to be obnoxiously lobotomized.

An uncensored qwen3.5/3.6 is more fun