Is llm-surgery-dark-arts-gpt-oss-60b-96a12 or DeepSeek-V4-Flash smaller?

Both publish quantizations; compare the table for the exact sizes.

Which uses less memory at long context?

DeepSeek-V4-Flash — its KV cache at 32K context is 12.2x smaller, because of how its attention layers cache.

Model comparison

llm-surgery-dark-arts-gpt-oss-60b-96a12 vs DeepSeek-V4-Flash

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: DeepSeek-V4-Flash's KV cache at 32K is 12.2× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	llm-surgery-dark-arts-gpt-oss-60b-96a12	DeepSeek-V4-Flash
Parameters	60.9B	291B
Architecture	gpt-oss	deepseek4
Layers	24	43
Native context	131,072	1,048,576
Mixture of experts	yes, 96 experts	yes, 256 experts
Quantizations published	34	12
Smallest quantization	31.28 GiB	76.87 GiB
Q4_K_M	41.48 GiB	—
Licence	apache-2.0	mit

KV cache by context

the term that decides long-context viability

Context	llm-surgery-dark-arts-gpt-oss-60b-96a12	DeepSeek-V4-Flash	Ratio
4,096	0.11 GiB	0.06 GiB	1.77×
8,192	0.21 GiB	0.06 GiB	3.26×
16,384	0.39 GiB	0.06 GiB	6.23×
32,768	0.77 GiB	0.06 GiB	12.19×
65,536	1.52 GiB	0.06 GiB	24.09×
131,072	3.02 GiB	0.06 GiB	47.91×