This is what it looks like to run Qwen 3.8 27B on a 5090 and get more than 100 tok/s.
6-month-old frontier AI, now running locally on a gaming GPU.
Claude is a narc and downgrades you all the time.
Meanwhile my own computer purrs away writing code with on-device models. Using Alibaba_Qwen Qwen 3.8 obliterated for some cyber work 🤓
This is not a one shot or single prompt.
I've been working with Qwen 3.8 27B local today, prioritizing fixes to clone a js water simulation. The water physics are the point, and they aren't even close to the reference... but they finally look alright.
Now check this out and compare yourself
Will share that via web with the entire prompt and comparison to models like GLM and Grok so you have a side by side view this is Qwen 3.8 27B 4bit Optimized Speed FP 16 created in Grok Build CLI running via mtplx, 132k context total
You have reached the end of the archive
All of qwen38