Is HarmonicHarlequin_v5-20B or Qwen3-30B-A3B-Thinking-2507 smaller?

Qwen3-30B-A3B-Thinking-2507 is smaller at Q4_K_M: 17.28 GiB against 18.71 GiB.

Which uses less memory at long context?

Qwen3-30B-A3B-Thinking-2507 — its KV cache at 32K context is 21.7x smaller, because of how its attention layers cache.

Model comparison

HarmonicHarlequin_v5-20B vs Qwen3-30B-A3B-Thinking-2507

At Q4_K_M, Qwen3-30B-A3B-Thinking-2507 is the smaller download — 18,556,685,824 bytes against 20,087,800,544. At long context the gap widens: Qwen3-30B-A3B-Thinking-2507's KV cache at 32K is 21.7× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	HarmonicHarlequin_v5-20B	Qwen3-30B-A3B-Thinking-2507
Parameters	33.3B	30.5B
Architecture	llama	qwen3moe
Layers	104	48
Native context	4,096	262,144
Mixture of experts	no	yes, 128 experts
Quantizations published	35	51
Smallest quantization	6.77 GiB	7.05 GiB
Q4_K_M	18.71 GiB	17.28 GiB
Licence	—	apache-2.0

KV cache by context

the term that decides long-context viability

Context	HarmonicHarlequin_v5-20B	Qwen3-30B-A3B-Thinking-2507	Ratio
4,096	8.13 GiB	0.38 GiB	21.67×
8,192	16.25 GiB	0.75 GiB	21.67×
16,384	32.50 GiB	1.50 GiB	21.67×
32,768	65.00 GiB	3.00 GiB	21.67×
65,536	130.00 GiB	6.00 GiB	21.67×
131,072	260.00 GiB	12.00 GiB	21.67×