I’m waiting for the local coding agent distillations that us mortals can run on our systems.
27B and Qwen flash next can run with somewhat realistic specs. I run 27B on my 7900 XTX.
24gb, luxury 😉 I have to make do with 8gb if vram but the previous 3.5-9b-artimus worked ok. I wonder how much of the 27B is used for the multi model support?
I offload mmproj to the CPU to keep as much model weights in VRAM as possible 🙂
I was lucky enough to get one a few months ago for under $1k before the price doubled. It was the last sanely priced 24GB card, but that didn’t last long. 😭
Same thing, I bought mine last month!
I’ve found that mimo is pretty okay for its size too
Seems the hype train is on its way. But I must say I’d rather read about results than have advertisement forwarded to me.
Weights or riot
Both!
This is the only set of benchmarks I can find at the moment: https://offloop.org/learn/qwen-4-max-preview-test. No idea if it’s fake slop or not, but it shows an early preview of Qwen 4 27B as being similar to Sonnet 5.5 on XHigh. Qwen 27B seems to be keeping pace with Sonnet, so this sounds plausible.
it’s so close, I’m gonna Qwum!


