Is gemma-4-12B-it-qat-q4_0-unquantized or Ministral-3-8B-Instruct-2512-BF16 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 1.7x smaller, because of how its attention layers cache.

Model comparison

gemma-4-12B-it-qat-q4_0-unquantized vs Ministral-3-8B-Instruct-2512-BF16

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: gemma-4-12B-it-qat-q4_0-unquantized's KV cache at 32K is 1.7× 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	Ministral-3-8B-Instruct-2512-BF16
Parameters	12.0B	8.9B
Architecture	gemma4	mistral3
Layers	48	34
Native context	262,144	262,144
Mixture of experts	no	no
Quantizations published	2	45
Smallest quantization	6.50 GiB	2.89 GiB
Q4_K_M	—	4.84 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	Ministral-3-8B-Instruct-2512-BF16	Ratio
4,096	0.72 GiB	0.53 GiB	1.35×
8,192	0.97 GiB	1.06 GiB	1.10×
16,384	1.47 GiB	2.13 GiB	1.45×
32,768	2.47 GiB	4.25 GiB	1.72×
65,536	4.47 GiB	8.50 GiB	1.90×
131,072	8.47 GiB	17.00 GiB	2.01×