Qwen 3.6 27B 30GB Same top p: 98.358 ± 0.033 % vs UD Q8 K XL 33GB Same top p: 97.426 ± 0.041 %

In three linesCustom quantization experiment on Qwen 3.6 27B: BF16→Q8_0 conversion targeting high-variance layers. Q8-CC model (30.47 GiB) achieves 98.358% vs UD Q8_K_XL (33.31 GiB) at 97.426% on wiki.test.raw. Mean KLD: 0.011324 vs 0.012100. Preliminary results without real-world performance benchmarks.

Summary generated by Claude — human-verified