Model comparison

Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensored vs Qwen3-30B-A3B-Thinking-2507

At Q4_K_M, Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensored is the smaller download — 16,540,273,504 bytes against 18,556,685,824. At long context the gap widens: Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensored's KV cache at 32K is 1.5× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensoredQwen3-30B-A3B-Thinking-2507
Parameters27.4B30.5B
Architectureqwen35qwen3moe
Layers6448
Native context262,144262,144
Mixture of expertsnoyes, 128 experts
Quantizations published3451
Smallest quantization5.80 GiB7.05 GiB
Q4_K_M15.40 GiB17.28 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextQwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensoredQwen3-30B-A3B-Thinking-2507Ratio
4,0960.25 GiB0.38 GiB1.50×
8,1920.50 GiB0.75 GiB1.50×
16,3841.00 GiB1.50 GiB1.50×
32,7682.00 GiB3.00 GiB1.50×
65,5364.00 GiB6.00 GiB1.50×
131,0728.00 GiB12.00 GiB1.50×