Is internlm2-math-plus-20b or Qwen3-30B-A3B-Thinking-2507 smaller?

internlm2-math-plus-20b is smaller at Q4_K_M: 11.16 GiB against 17.28 GiB.

Which uses less memory at long context?

Qwen3-30B-A3B-Thinking-2507 — its KV cache at 32K context is 2.0x smaller, because of how its attention layers cache.

Model comparison

internlm2-math-plus-20b vs Qwen3-30B-A3B-Thinking-2507

At Q4_K_M, internlm2-math-plus-20b is the smaller download — 11,984,469,440 bytes against 18,556,685,824. At long context the gap widens: Qwen3-30B-A3B-Thinking-2507's KV cache at 32K is 2.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	internlm2-math-plus-20b	Qwen3-30B-A3B-Thinking-2507
Parameters	19.9B	30.5B
Architecture	internlm2	qwen3moe
Layers	48	48
Native context	8,192	262,144
Mixture of experts	no	yes, 128 experts
Quantizations published	68	51
Smallest quantization	4.23 GiB	7.05 GiB
Q4_K_M	11.16 GiB	17.28 GiB
Licence	other	apache-2.0

KV cache by context

the term that decides long-context viability

Context	internlm2-math-plus-20b	Qwen3-30B-A3B-Thinking-2507	Ratio
4,096	0.75 GiB	0.38 GiB	2.00×
8,192	1.50 GiB	0.75 GiB	2.00×
16,384	3.00 GiB	1.50 GiB	2.00×
32,768	6.00 GiB	3.00 GiB	2.00×
65,536	12.00 GiB	6.00 GiB	2.00×
131,072	24.00 GiB	12.00 GiB	2.00×