Qwen3.8-Flash-Next -- feedback beyond benchmarks?
So, I finally got this thing running on my local hardware, but it takes most of the resources of my box. Before I throw away the desktop/etc aspect of my hardware and relegate this to a dedicated inference node... is there anyone actually *using* this model for agentic coding? How's it perform? I could go with qwen 3.8 27b dense using other hardware, and leave this open as my workstation plus smaller classifier models, etc. Is 3.8-flash-next good enough to pretty much dedicate this box to? What have you made with it?