Hey guys! Today, we’re releasing the highest accuracy Qwen3.8-27B GGUFs.
Our new Unsloth AI Dynamic V3.1 GGUFs lead by 5–10% on average across Terminal Bench, DeepSWE & more benchmarks. We’re working on even better version of Dynamic quants for Qwen4
Run on 12GB+ RAM setups.
To compare GSQ-RCO and Exllama-v3, we had to exclude the token embedding layer from the x axis, as Exllama-v3 uses BF16 for it. If you consider the full download size and total RAM usage for Exllama-v3, then adding the BF16 embedding table and including all embedding sizes changes the plots. Our main plots’s x axis removes the embedding layer sizes from all quant providers. However on pure RAM usage and disk space usage vs accuracy, the below plot in the comments the true reflection of actual usage.
From their Reddit post.
Anyone know which variant would work best with 32GB RAM / 16GB VRAM?
Hugging Face has some kind of hardware compatibility feature, but I’m trying to avoid making yet another account to use it.man if they could have just asked the AI if anyone with colorblindness could read these fucking charts.
I’m gonna try this, but I have a feeling my machine will be too slow to run it.
What are your specs?
I’ve got an AMD Ryzen 5 5600 and 32 GB of RAM, so I had to settle for Qwen3.8-27B-GGUF:UD-Q4_K_M. It runs, but the reality is that it’s too slow for me to really use.
But I still appreciate that it exists! :)
