Which uses less memory at long context?

Qwen3.6-27B — its KV cache at 32K context is 2.2x smaller, because of how its attention layers cache.

Model comparison

Qwen3.6-27B vs Qwen3-48B-A4B-Savant-Commander-Distill-12X-Closed-Open-Heretic-Uncensored

Q: Is Qwen3.6-27B or Qwen3-48B-A4B-Savant-Commander-Distill-12X-Closed-Open-Heretic-Uncensored smaller?

Qwen3.6-27B is smaller at Q4_K_M: 15.41 GiB against 19.01 GiB.

At Q4_K_M, Qwen3.6-27B is the smaller download — 16,547,397,888 bytes against 20,414,848,064. At long context the gap widens: Qwen3.6-27B's KV cache at 32K is 2.2× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Qwen3.6-27B	Qwen3-48B-A4B-Savant-Commander-Distill-12X-Closed-Open-Heretic-Uncensored
Parameters	27.8B	33.6B
Architecture	qwen35	qwen3moe
Layers	64	36
Native context	262,144	262,144
Mixture of experts	no	yes, 12 experts
Quantizations published	40	39
Smallest quantization	8.74 GiB	6.50 GiB
Q4_K_M	15.41 GiB	19.01 GiB
Licence	apache-2.0	apache-2.0

KV cache by context

the term that decides long-context viability

Context	Qwen3.6-27B	Qwen3-48B-A4B-Savant-Commander-Distill-12X-Closed-Open-Heretic-Uncensored	Ratio
4,096	0.25 GiB	0.56 GiB	2.25×
8,192	0.50 GiB	1.13 GiB	2.25×
16,384	1.00 GiB	2.25 GiB	2.25×
32,768	2.00 GiB	4.50 GiB	2.25×
65,536	4.00 GiB	9.00 GiB	2.25×
131,072	8.00 GiB	18.00 GiB	2.25×