Is gemma-4-26B-A4B-it or gemma-3-12b-it-qat-int4-unquantized smaller?

gemma-3-12b-it-qat-int4-unquantized is smaller at Q4_K_M: 6.80 GiB against 15.64 GiB.

Which uses less memory at long context?

We cannot resolve the KV layout for one of these models, so we do not publish a comparison.

Model comparison

At Q4_K_M, gemma-3-12b-it-qat-int4-unquantized is the smaller download — 7,300,778,848 bytes against 16,796,017,248.

From the file· summed bytes, KV per layer

	gemma-4-26B-A4B-it	gemma-3-12b-it-qat-int4-unquantized
Parameters	26.5B	12.2B
Architecture	gemma4	gemma3
Layers	30	—
Native context	262,144	—
Mixture of experts	yes, 128 experts	no
Quantizations published	46	15
Smallest quantization	8.99 GiB	4.11 GiB
Q4_K_M	15.64 GiB	6.80 GiB
Licence	apache-2.0	gemma

the term that decides long-context viability

Context	gemma-4-26B-A4B-it	gemma-3-12b-it-qat-int4-unquantized	Ratio
4,096	0.45 GiB	—	—
8,192	0.61 GiB	—	—
16,384	0.92 GiB	—	—
32,768	1.54 GiB	—	—
65,536	2.79 GiB	—	—
131,072	5.29 GiB	—	—