Is GLM-4.5-Air-Derestricted or Hy3 smaller?

GLM-4.5-Air-Derestricted is smaller at Q4_K_M: 68.45 GiB against 168.57 GiB.

Which uses less memory at long context?

GLM-4.5-Air-Derestricted — its KV cache at 32K context is 1.7x smaller, because of how its attention layers cache.

Model comparison

GLM-4.5-Air-Derestricted vs Hy3

At Q4_K_M, GLM-4.5-Air-Derestricted is the smaller download — 73,501,936,704 bytes against 180,999,059,328. At long context the gap widens: GLM-4.5-Air-Derestricted's KV cache at 32K is 1.7× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	GLM-4.5-Air-Derestricted	Hy3
Parameters	110B	299B
Architecture	glm4moe	hy_v3
Layers	46	80
Native context	131,072	262,144
Mixture of experts	yes, 128 experts	yes, 192 experts
Quantizations published	48	37
Smallest quantization	35.26 GiB	59.65 GiB
Q4_K_M	68.45 GiB	168.57 GiB
Licence	mit	apache-2.0

KV cache by context

the term that decides long-context viability

Context	GLM-4.5-Air-Derestricted	Hy3	Ratio
4,096	0.72 GiB	1.25 GiB	1.74×
8,192	1.44 GiB	2.50 GiB	1.74×
16,384	2.88 GiB	5.00 GiB	1.74×
32,768	5.75 GiB	10.00 GiB	1.74×
65,536	11.50 GiB	20.00 GiB	1.74×
131,072	23.00 GiB	40.00 GiB	1.74×