Is Huihui-gpt-oss-20b-BF16-abliterated or Qwen3-Coder-30B-A3B-Instruct smaller?

Huihui-gpt-oss-20b-BF16-abliterated is smaller at Q4_K_M: 10.87 GiB against 17.28 GiB.

Which uses less memory at long context?

Huihui-gpt-oss-20b-BF16-abliterated — its KV cache at 32K context is 3.9x smaller, because of how its attention layers cache.

Model comparison

Huihui-gpt-oss-20b-BF16-abliterated vs Qwen3-Coder-30B-A3B-Instruct

At Q4_K_M, Huihui-gpt-oss-20b-BF16-abliterated is the smaller download — 11,673,418,400 bytes against 18,556,689,568. At long context the gap widens: Huihui-gpt-oss-20b-BF16-abliterated's KV cache at 32K is 3.9× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Huihui-gpt-oss-20b-BF16-abliterated	Qwen3-Coder-30B-A3B-Instruct
Parameters	20.9B	30.5B
Architecture	gpt-oss	qwen3moe
Layers	24	48
Native context	131,072	262,144
Mixture of experts	yes, 32 experts	yes, 128 experts
Quantizations published	33	46
Smallest quantization	10.70 GiB	7.46 GiB
Q4_K_M	10.87 GiB	17.28 GiB
Licence	apache-2.0	apache-2.0

KV cache by context

the term that decides long-context viability

Context	Huihui-gpt-oss-20b-BF16-abliterated	Qwen3-Coder-30B-A3B-Instruct	Ratio
4,096	0.11 GiB	0.38 GiB	3.37×
8,192	0.21 GiB	0.75 GiB	3.66×
16,384	0.39 GiB	1.50 GiB	3.82×
32,768	0.77 GiB	3.00 GiB	3.91×
65,536	1.52 GiB	6.00 GiB	3.95×
131,072	3.02 GiB	12.00 GiB	3.98×