Is Qwen3-Coder-30B-A3B-Instruct or Yi-34B-200K-DARE-megamerge-v8 smaller?

Qwen3-Coder-30B-A3B-Instruct is smaller at Q4_K_M: 17.28 GiB against 19.24 GiB.

Which uses less memory at long context?

Qwen3-Coder-30B-A3B-Instruct — its KV cache at 32K context is 2.5x smaller, because of how its attention layers cache.

Model comparison

Qwen3-Coder-30B-A3B-Instruct vs Yi-34B-200K-DARE-megamerge-v8

At Q4_K_M, Qwen3-Coder-30B-A3B-Instruct is the smaller download — 18,556,689,568 bytes against 20,658,730,400. At long context the gap widens: Qwen3-Coder-30B-A3B-Instruct's KV cache at 32K is 2.5× 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	Yi-34B-200K-DARE-megamerge-v8
Parameters	30.5B	34.4B
Architecture	qwen3moe	llama
Layers	48	60
Native context	262,144	200,000
Mixture of experts	yes, 128 experts	no
Quantizations published	46	50
Smallest quantization	7.46 GiB	6.98 GiB
Q4_K_M	17.28 GiB	19.24 GiB
Licence	apache-2.0	other

KV cache by context

the term that decides long-context viability

Context	Qwen3-Coder-30B-A3B-Instruct	Yi-34B-200K-DARE-megamerge-v8	Ratio
4,096	0.38 GiB	0.94 GiB	2.50×
8,192	0.75 GiB	1.88 GiB	2.50×
16,384	1.50 GiB	3.75 GiB	2.50×
32,768	3.00 GiB	7.50 GiB	2.50×
65,536	6.00 GiB	15.00 GiB	2.50×
131,072	12.00 GiB	30.00 GiB	2.50×