Is gemma-4-E4B-it-QAT-SOMPOA-heresy or Qwen3-8B smaller?

Qwen3-8B is smaller at Q4_K_M: 4.68 GiB against 4.94 GiB.

Which uses less memory at long context?

gemma-4-E4B-it-QAT-SOMPOA-heresy — its KV cache at 32K context is 8.9x smaller, because of how its attention layers cache.

Model comparison

gemma-4-E4B-it-QAT-SOMPOA-heresy vs Qwen3-8B

At Q4_K_M, Qwen3-8B is the smaller download — 5,027,783,488 bytes against 5,302,273,248. At long context the gap widens: gemma-4-E4B-it-QAT-SOMPOA-heresy's KV cache at 32K is 8.9× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	gemma-4-E4B-it-QAT-SOMPOA-heresy	Qwen3-8B
Parameters	7.9B	8.2B
Architecture	gemma4	qwen3
Layers	42	36
Native context	131,072	40,960
Mixture of experts	no	no
Quantizations published	36	51
Smallest quantization	3.06 GiB	2.12 GiB
Q4_K_M	4.94 GiB	4.68 GiB
Licence	apache-2.0	apache-2.0

KV cache by context

the term that decides long-context viability

Context	gemma-4-E4B-it-QAT-SOMPOA-heresy	Qwen3-8B	Ratio
4,096	0.12 GiB	0.56 GiB	4.57×
8,192	0.18 GiB	1.13 GiB	6.33×
16,384	0.29 GiB	2.25 GiB	7.84×
32,768	0.51 GiB	4.50 GiB	8.90×
65,536	0.94 GiB	9.00 GiB	9.54×
131,072	1.82 GiB	18.00 GiB	9.90×