I just successfully completed the coolest thing I've done so far with AI using Qwen 3.8 27B NVFP4 for NInfer.
I had it reverse engineer Starlight Precise 2.6, the best video upscaler so I can run it natively in Linux. It was previously being used inside Wine. This seems to have
The best 4 results head to head.
Fable Sol Kimi k3 Qwen 3.8 max
What can a local coding agent build from ONE prompt?
Apparently this. 👇 NousResearch Hermes Agent + Alibaba_Qwen Qwen 3.8 27B Dense + MTP speculative decoding on a single RTX 3090 (credit to UnslothAI for the fantastic IQ4_XS quant — making this level of local AI
Atomic Chat HQ ran an identical prompt suite across four quantization tiers of Qwen 3.8 27B to evaluate voxel island generation.
Their benchmark covered complex environments, including cliffside villages, autumn roads, and steampunk sky islands. The tech community expected
You have reached the end of the archive
All of qwen38