Which uses less memory at long context?

Gemma-3-4B-VL-it-Gemini-Pro-Heretic-Uncensored-Thinking — its KV cache at 32K context is 1.3x smaller, because of how its attention layers cache.

Model comparison

Qwen3.5-9B vs Gemma-3-4B-VL-it-Gemini-Pro-Heretic-Uncensored-Thinking

Q: Is Qwen3.5-9B or Gemma-3-4B-VL-it-Gemini-Pro-Heretic-Uncensored-Thinking smaller?

Gemma-3-4B-VL-it-Gemini-Pro-Heretic-Uncensored-Thinking is smaller at Q4_K_M: 2.32 GiB against 5.24 GiB.

At Q4_K_M, Gemma-3-4B-VL-it-Gemini-Pro-Heretic-Uncensored-Thinking is the smaller download — 2,489,895,200 bytes against 5,627,044,256. At long context the gap widens: Gemma-3-4B-VL-it-Gemini-Pro-Heretic-Uncensored-Thinking's KV cache at 32K is 1.3× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Qwen3.5-9B	Gemma-3-4B-VL-it-Gemini-Pro-Heretic-Uncensored-Thinking
Parameters	9.7B	4.3B
Architecture	qwen35	gemma3
Layers	32	34
Native context	262,144	131,072
Mixture of experts	no	no
Quantizations published	34	12
Smallest quantization	2.97 GiB	1.61 GiB
Q4_K_M	5.24 GiB	2.32 GiB
Licence	apache-2.0	apache-2.0

KV cache by context

the term that decides long-context viability

Context	Qwen3.5-9B	Gemma-3-4B-VL-it-Gemini-Pro-Heretic-Uncensored-Thinking	Ratio
4,096	0.13 GiB	0.25 GiB	1.98×
8,192	0.25 GiB	0.33 GiB	1.30×
16,384	0.50 GiB	0.48 GiB	1.04×
32,768	1.00 GiB	0.79 GiB	1.26×
65,536	2.00 GiB	1.42 GiB	1.41×
131,072	4.00 GiB	2.67 GiB	1.50×