Is gemma-4-12B-it-qat-q4_0-unquantized or llm-jp-4-32b-a3b-thinking smaller?

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

Which uses less memory at long context?

llm-jp-4-32b-a3b-thinking — its KV cache at 32K context is 1.2x smaller, because of how its attention layers cache.

Model comparison

gemma-4-12B-it-qat-q4_0-unquantized vs llm-jp-4-32b-a3b-thinking

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: llm-jp-4-32b-a3b-thinking's KV cache at 32K is 1.2× 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	llm-jp-4-32b-a3b-thinking
Parameters	12.0B	32.1B
Architecture	gemma4	qwen3moe
Layers	48	32
Native context	262,144	65,536
Mixture of experts	no	yes, 128 experts
Quantizations published	2	11
Smallest quantization	6.50 GiB	15.60 GiB
Q4_K_M	—	19.93 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	llm-jp-4-32b-a3b-thinking	Ratio
4,096	0.72 GiB	0.25 GiB	2.88×
8,192	0.97 GiB	0.50 GiB	1.94×
16,384	1.47 GiB	1.00 GiB	1.47×
32,768	2.47 GiB	2.00 GiB	1.23×
65,536	4.47 GiB	4.00 GiB	1.12×
131,072	8.47 GiB	8.00 GiB	1.06×