Is DeepSeek-OCR or Qwen3-8B smaller?

DeepSeek-OCR is smaller at Q4_K_M: 1.82 GiB against 4.68 GiB.

Which uses less memory at long context?

DeepSeek-OCR — its KV cache at 32K context is 4.8x smaller, because of how its attention layers cache.

Model comparison

DeepSeek-OCR vs Qwen3-8B

At Q4_K_M, DeepSeek-OCR is the smaller download — 1,950,326,592 bytes against 5,027,783,488. At long context the gap widens: DeepSeek-OCR's KV cache at 32K is 4.8× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	DeepSeek-OCR	Qwen3-8B
Parameters	3.3B	8.2B
Architecture	deepseek_vl_v2	qwen3
Layers	12	36
Native context	8,192	40,960
Mixture of experts	yes, 64 experts	no
Quantizations published	11	51
Smallest quantization	1.54 GiB	2.12 GiB
Q4_K_M	1.82 GiB	4.68 GiB
Licence	—	apache-2.0

KV cache by context

the term that decides long-context viability

Context	DeepSeek-OCR	Qwen3-8B	Ratio
4,096	0.12 GiB	0.56 GiB	4.80×
8,192	0.23 GiB	1.13 GiB	4.80×
16,384	0.47 GiB	2.25 GiB	4.80×
32,768	0.94 GiB	4.50 GiB	4.80×
65,536	1.88 GiB	9.00 GiB	4.80×
131,072	3.75 GiB	18.00 GiB	4.80×