Is gemma-4-12B-it-qat-q4_0-unquantized or Qwen3.5-9B-Nikusui-v1 smaller?

Both publish quantizations; compare the table for the exact sizes.

Which uses less memory at long context?

Qwen3.5-9B-Nikusui-v1 — its KV cache at 32K context is 2.5x smaller, because of how its attention layers cache.

Model comparison

gemma-4-12B-it-qat-q4_0-unquantized vs Qwen3.5-9B-Nikusui-v1

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Qwen3.5-9B-Nikusui-v1's KV cache at 32K is 2.5× 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-qat-q4_0-unquantized	Qwen3.5-9B-Nikusui-v1
Parameters	12.0B	9.7B
Architecture	gemma4	qwen35
Layers	48	32
Native context	262,144	262,144
Mixture of experts	no	no
Quantizations published	2	29
Smallest quantization	6.50 GiB	3.65 GiB
Q4_K_M	—	5.38 GiB
Licence	apache-2.0	apache-2.0

KV cache by context

the term that decides long-context viability

Context	gemma-4-12B-it-qat-q4_0-unquantized	Qwen3.5-9B-Nikusui-v1	Ratio
4,096	0.72 GiB	0.13 GiB	5.75×
8,192	0.97 GiB	0.25 GiB	3.87×
16,384	1.47 GiB	0.50 GiB	2.94×
32,768	2.47 GiB	1.00 GiB	2.47×
65,536	4.47 GiB	2.00 GiB	2.23×
131,072	8.47 GiB	4.00 GiB	2.12×