AnalysisAI ModelsAugust 28, 2026

SOTA GGUFs released for Qwen3.8-27B at 2.5-3.0 bpw

Community release quantizes Qwen3.8-27B with GSQ (Gumbel-Softmax Quantization) plus RCO, claiming higher quality at the same file size across 2.5 to 3.0 bits per weight. Both the search and the quantizer are learned jointly.

1 source

More stories today

Open the live feed