Is Vikhr-Gemma-2B-instruct or Qwen3-4B smaller?

Vikhr-Gemma-2B-instruct is smaller at Q4_K_M: 1.59 GiB against 2.33 GiB.

Which uses less memory at long context?

Vikhr-Gemma-2B-instruct — its KV cache at 32K context is 2.4x smaller, because of how its attention layers cache.

Model comparison

Vikhr-Gemma-2B-instruct vs Qwen3-4B

At Q4_K_M, Vikhr-Gemma-2B-instruct is the smaller download — 1,708,582,848 bytes against 2,497,280,256. At long context the gap widens: Vikhr-Gemma-2B-instruct's KV cache at 32K is 2.4× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Vikhr-Gemma-2B-instruct	Qwen3-4B
Parameters	2.6B	4.0B
Architecture	gemma2	qwen3
Layers	26	36
Native context	8,192	40,960
Mixture of experts	no	no
Quantizations published	46	29
Smallest quantization	0.78 GiB	1.01 GiB
Q4_K_M	1.59 GiB	2.33 GiB
Licence	apache-2.0	—

KV cache by context

the term that decides long-context viability

Context	Vikhr-Gemma-2B-instruct	Qwen3-4B	Ratio
4,096	0.41 GiB	0.56 GiB	1.38×
8,192	0.63 GiB	1.13 GiB	1.77×
16,384	1.04 GiB	2.25 GiB	2.16×
32,768	1.85 GiB	4.50 GiB	2.43×
65,536	3.48 GiB	9.00 GiB	2.59×
131,072	6.73 GiB	18.00 GiB	2.68×