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

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

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

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

KV cache by context

the term that decides long-context viability

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