Is gemma-4-12B-it or Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness smaller?

Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness is smaller at Q4_K_M: 5.09 GiB against 6.63 GiB.

Which uses less memory at long context?

Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness — its KV cache at 32K context is 4.9x smaller, because of how its attention layers cache.

Model comparison

gemma-4-12B-it vs Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness

At Q4_K_M, Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness is the smaller download — 5,463,652,192 bytes against 7,121,861,440. At long context the gap widens: Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness's KV cache at 32K is 4.9× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	gemma-4-12B-it	Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness
Parameters	12.0B	9.1B
Architecture	gemma4	qwen35
Layers	48	16
Native context	262,144	262,144
Mixture of experts	no	no
Quantizations published	59	2
Smallest quantization	3.92 GiB	4.88 GiB
Q4_K_M	6.63 GiB	5.09 GiB
Licence	apache-2.0	apache-2.0

KV cache by context

the term that decides long-context viability

Context	gemma-4-12B-it	Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness	Ratio
4,096	0.72 GiB	0.06 GiB	11.49×
8,192	0.97 GiB	0.13 GiB	7.75×
16,384	1.47 GiB	0.25 GiB	5.87×
32,768	2.47 GiB	0.50 GiB	4.94×
65,536	4.47 GiB	1.00 GiB	4.47×
131,072	8.47 GiB	2.00 GiB	4.23×