Is Qwen3-Coder-30B-A3B-Instruct or ERNIE-21B-A3B-Claude-4.5-High-OPUS-Thinking smaller?

ERNIE-21B-A3B-Claude-4.5-High-OPUS-Thinking is smaller at Q4_K_M: 12.34 GiB against 17.28 GiB.

Which uses less memory at long context?

ERNIE-21B-A3B-Claude-4.5-High-OPUS-Thinking — its KV cache at 32K context is 1.7x smaller, because of how its attention layers cache.

Model comparison

Qwen3-Coder-30B-A3B-Instruct vs ERNIE-21B-A3B-Claude-4.5-High-OPUS-Thinking

At Q4_K_M, ERNIE-21B-A3B-Claude-4.5-High-OPUS-Thinking is the smaller download — 13,245,832,896 bytes against 18,556,689,568. At long context the gap widens: ERNIE-21B-A3B-Claude-4.5-High-OPUS-Thinking'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

	Qwen3-Coder-30B-A3B-Instruct	ERNIE-21B-A3B-Claude-4.5-High-OPUS-Thinking
Parameters	30.5B	21.8B
Architecture	qwen3moe	ernie4_5-moe
Layers	48	28
Native context	262,144	131,072
Mixture of experts	yes, 128 experts	no
Quantizations published	46	34
Smallest quantization	7.46 GiB	4.17 GiB
Q4_K_M	17.28 GiB	12.34 GiB
Licence	apache-2.0	apache-2.0

KV cache by context

the term that decides long-context viability

Context	Qwen3-Coder-30B-A3B-Instruct	ERNIE-21B-A3B-Claude-4.5-High-OPUS-Thinking	Ratio
4,096	0.38 GiB	0.22 GiB	1.71×
8,192	0.75 GiB	0.44 GiB	1.71×
16,384	1.50 GiB	0.88 GiB	1.71×
32,768	3.00 GiB	1.75 GiB	1.71×
65,536	6.00 GiB	3.50 GiB	1.71×
131,072	12.00 GiB	7.00 GiB	1.71×