There is a high quality Qwen3.8-27B out there for 12 - 16GB folks, it's @Eschalabs Qwen3.8-27B-W2, has really good quality for it's size.
This will fit on 16GB fine, q8_0 quant should fit on 12GB but with limited context.
It's their work, i only worked on CUDA for us llama.cpp folks and created quant. Needs my "llama.cpp-escha" fork to work
https://t.co/kDUfc5KvDT