Is GLM-OCR or Qwen3-1.7B smaller?

GLM-OCR is smaller at Q4_K_M: 0.51 GiB against 1.03 GiB.

Which uses less memory at long context?

GLM-OCR — its KV cache at 32K context is 1.8x smaller, because of how its attention layers cache.

Model comparison

GLM-OCR vs Qwen3-1.7B

At Q4_K_M, GLM-OCR is the smaller download — 548,515,136 bytes against 1,107,409,472. At long context the gap widens: GLM-OCR's KV cache at 32K is 1.8× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	GLM-OCR	Qwen3-1.7B
Parameters	1.3B	2.0B
Architecture	glm4	qwen3
Layers	16	28
Native context	131,072	40,960
Mixture of experts	no	no
Quantizations published	30	48
Smallest quantization	0.34 GiB	0.50 GiB
Q4_K_M	0.51 GiB	1.03 GiB
Licence	—	—

KV cache by context

the term that decides long-context viability

Context	GLM-OCR	Qwen3-1.7B	Ratio
4,096	0.25 GiB	0.44 GiB	1.75×
8,192	0.50 GiB	0.88 GiB	1.75×
16,384	1.00 GiB	1.75 GiB	1.75×
32,768	2.00 GiB	3.50 GiB	1.75×
65,536	4.00 GiB	7.00 GiB	1.75×
131,072	8.00 GiB	14.00 GiB	1.75×