I think I’ve found my new daily driver: Qwen 3.8 27B Q4_K_M.
Running locally at around 42 tok/s decode on a pretty modest setup: • i7-6700T • 64 GB RAM • 2× RTX 3060 12 GB It’s fast enough to stay out of the way and capable enough that I keep reaching for it over the other
I got 200% speed boost in Qwen 3.8 Flash Next, it's now 28-24tok/sec in 64gb of ram ddr4 and 3090.
It was around 11 tok/s when I first started, I bet ddr5 is going to be faster. Cafe-llama.cpp it's not just one more fork.
A free 27B Chinese AI model can now analyze your SEO research, videos, documents, workflows, and code.
But the interesting part isn't that it's free. It's what happens when you give Qwen 3.8-27B your entire business context.
Qwen 3.8 unhinged is actually more wikd than I expected 😳
You have reached the end of the archive
All of qwen38