If you write a script that can alternately load and offload Qwen 3.8's uncensored model and Minimax H3
If you write a script that can alternately load and offload Qwen 3.8's uncensored model and Minimax H3, you can generate unlimited videos completely locally. Even one machine. I no longer have to change the prompt every now and then on the comfy web screen. Look at the image and see what's wrong, fix it.
I'm blown away by Qwen 3.8 27B.
A few months ago when I was testing open-weight models, it seemed clear that only GLM and Kimi class models (+1T parameters) could get close to GPT-5.6 / Opus / Fable levels. And now this tiny 27B model gets almost as close. I gotta say, China
Even wat testen met qwen 3.8 27b op mijn RTX 5090 laptop (24gb)
Resultaten per test: - Q4 D3.0 · MTP · thinking off — 73,1 t/s, ctx 187K - Q4 D3.0 · MTP · thinking on — 45,9 t/s - Q3 Unleashed · MTP · thinking on — 56,5 t/s, ctx 131K - Q4 D3.0 · DFlash2 · thinking off —
Qwen 3.8 Max 0902 just went live, and this AI worked alone for 10 days straight.
Alibaba gave it an empty folder and said "build a working tool." No starter code. No human help. Nothing. → It set up its own job queue → Ran its own tests → Caught its own bugs and fixed
You have reached the end of the archive
All of qwen38