Is gemma-3-12b-it-abliterated or gemma-4-31B-it smaller?

gemma-3-12b-it-abliterated is smaller at Q4_K_M: 6.80 GiB against 17.07 GiB.

Which uses less memory at long context?

gemma-3-12b-it-abliterated — its KV cache at 32K context is 2.5x smaller, because of how its attention layers cache.

Model comparison

gemma-3-12b-it-abliterated vs gemma-4-31B-it

At Q4_K_M, gemma-3-12b-it-abliterated is the smaller download — 7,300,778,336 bytes against 18,323,733,440. At long context the gap widens: gemma-3-12b-it-abliterated'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-3-12b-it-abliterated	gemma-4-31B-it
Parameters	12.2B	31.3B
Architecture	gemma3	gemma4
Layers	48	60
Native context	131,072	262,144
Mixture of experts	no	no
Quantizations published	38	55
Smallest quantization	3.58 GiB	7.95 GiB
Q4_K_M	6.80 GiB	17.07 GiB
Licence	gemma	apache-2.0

KV cache by context

the term that decides long-context viability

Context	gemma-3-12b-it-abliterated	gemma-4-31B-it	Ratio
4,096	0.72 GiB	1.80 GiB	2.50×
8,192	0.97 GiB	2.42 GiB	2.50×
16,384	1.47 GiB	3.67 GiB	2.50×
32,768	2.47 GiB	6.17 GiB	2.50×
65,536	4.47 GiB	11.17 GiB	2.50×
131,072	8.47 GiB	21.17 GiB	2.50×