Is gemma-4-12B-it-heretic_decensored or gemma-4-12B-it-qat-q4_0-unquantized smaller?

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

Which uses less memory at long context?

gemma-4-12B-it-qat-q4_0-unquantized — its KV cache at 32K context is 1.0x smaller, because of how its attention layers cache.

Model comparison

These two publish different quantization sets; the table below has the exact sizes.

From the file· summed bytes, KV per layer

	gemma-4-12B-it-heretic_decensored	gemma-4-12B-it-qat-q4_0-unquantized
Parameters	12.0B	12.0B
Architecture	gemma4	gemma4
Layers	48	48
Native context	262,144	262,144
Mixture of experts	no	no
Quantizations published	49	2
Smallest quantization	2.78 GiB	6.50 GiB
Q4_K_M	6.87 GiB	—
Licence	apache-2.0	apache-2.0

the term that decides long-context viability

Context	gemma-4-12B-it-heretic_decensored	gemma-4-12B-it-qat-q4_0-unquantized	Ratio
4,096	0.72 GiB	0.72 GiB	1.00×
8,192	0.97 GiB	0.97 GiB	1.00×
16,384	1.47 GiB	1.47 GiB	1.00×
32,768	2.47 GiB	2.47 GiB	1.00×
65,536	4.47 GiB	4.47 GiB	1.00×
131,072	8.47 GiB	8.47 GiB	1.00×