Is GPT-NeoX-20B-Erebus 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 13.4x smaller, because of how its attention layers cache.

Model comparison

GPT-NeoX-20B-Erebus vs gemma-4-12B-it-qat-q4_0-unquantized

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 13.4× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	GPT-NeoX-20B-Erebus	gemma-4-12B-it-qat-q4_0-unquantized
Parameters	20.6B	12.0B
Architecture	gptneox	gemma4
Layers	44	48
Native context	2,048	262,144
Mixture of experts	no	no
Quantizations published	23	2
Smallest quantization	4.12 GiB	6.50 GiB
Q4_K_M	—	—
Licence	apache-2.0	apache-2.0

KV cache by context

the term that decides long-context viability

Context	GPT-NeoX-20B-Erebus	gemma-4-12B-it-qat-q4_0-unquantized	Ratio
4,096	4.13 GiB	0.72 GiB	5.74×
8,192	8.25 GiB	0.97 GiB	8.52×
16,384	16.50 GiB	1.47 GiB	11.23×
32,768	33.00 GiB	2.47 GiB	13.37×
65,536	66.00 GiB	4.47 GiB	14.77×
131,072	132.00 GiB	8.47 GiB	15.59×