Is gpt-oss-20b-heretic or Qwen3.8-27B smaller?

gpt-oss-20b-heretic is smaller at Q4_K_M: 14.72 GiB against 15.66 GiB.

Which uses less memory at long context?

gpt-oss-20b-heretic — its KV cache at 32K context is 2.6x smaller, because of how its attention layers cache.

Model comparison

gpt-oss-20b-heretic vs Qwen3.8-27B

At Q4_K_M, gpt-oss-20b-heretic is the smaller download — 15,805,137,056 bytes against 16,810,714,336. At long context the gap widens: gpt-oss-20b-heretic's KV cache at 32K is 2.6× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	gpt-oss-20b-heretic	Qwen3.8-27B
Parameters	20.9B	27.8B
Architecture	gpt-oss	qwen35
Layers	24	64
Native context	131,072	262,144
Mixture of experts	yes, 32 experts	no
Quantizations published	37	22
Smallest quantization	10.73 GiB	8.39 GiB
Q4_K_M	14.72 GiB	15.66 GiB
Licence	apache-2.0	apache-2.0

KV cache by context

the term that decides long-context viability

Context	gpt-oss-20b-heretic	Qwen3.8-27B	Ratio
4,096	0.11 GiB	0.25 GiB	2.25×
8,192	0.21 GiB	0.50 GiB	2.44×
16,384	0.39 GiB	1.00 GiB	2.55×
32,768	0.77 GiB	2.00 GiB	2.61×
65,536	1.52 GiB	4.00 GiB	2.64×
131,072	3.02 GiB	8.00 GiB	2.65×