Model comparison

Qwen3.5-9B-Star-Trek-TNG-DS9-Heretic-Uncensored-Thinking vs Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Qwen3.5-9B-Star-Trek-TNG-DS9-Heretic-Uncensored-Thinking's KV cache at 32K is 2.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

Qwen3.5-9B-Star-Trek-TNG-DS9-Heretic-Uncensored-ThinkingQwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP
Parameters9.4B27.8B
Architectureqwen35qwen35
Layers3264
Native context262,144262,144
Mixture of expertsnono
Quantizations published234
Smallest quantization9.77 GiB10.12 GiB
Q4_K_M—34.04 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextQwen3.5-9B-Star-Trek-TNG-DS9-Heretic-Uncensored-ThinkingQwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTPRatio
4,0960.13 GiB0.25 GiB2.00×
8,1920.25 GiB0.50 GiB2.00×
16,3840.50 GiB1.00 GiB2.00×
32,7681.00 GiB2.00 GiB2.00×
65,5362.00 GiB4.00 GiB2.00×
131,0724.00 GiB8.00 GiB2.00×