Is Qwen3-Coder-30B-A3B-Instruct or llm-jp-4-32b-a3b-thinking smaller?

Qwen3-Coder-30B-A3B-Instruct is smaller at Q4_K_M: 17.28 GiB against 19.93 GiB.

Which uses less memory at long context?

llm-jp-4-32b-a3b-thinking — its KV cache at 32K context is 1.5x smaller, because of how its attention layers cache.

Model comparison

Qwen3-Coder-30B-A3B-Instruct vs llm-jp-4-32b-a3b-thinking

At Q4_K_M, Qwen3-Coder-30B-A3B-Instruct is the smaller download — 18,556,689,568 bytes against 21,400,590,016. At long context the gap widens: llm-jp-4-32b-a3b-thinking's KV cache at 32K is 1.5× 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	llm-jp-4-32b-a3b-thinking
Parameters	30.5B	32.1B
Architecture	qwen3moe	qwen3moe
Layers	48	32
Native context	262,144	65,536
Mixture of experts	yes, 128 experts	yes, 128 experts
Quantizations published	46	11
Smallest quantization	7.46 GiB	15.60 GiB
Q4_K_M	17.28 GiB	19.93 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	llm-jp-4-32b-a3b-thinking	Ratio
4,096	0.38 GiB	0.25 GiB	1.50×
8,192	0.75 GiB	0.50 GiB	1.50×
16,384	1.50 GiB	1.00 GiB	1.50×
32,768	3.00 GiB	2.00 GiB	1.50×
65,536	6.00 GiB	4.00 GiB	1.50×
131,072	12.00 GiB	8.00 GiB	1.50×