Is Hermes-4-405B 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 250.0x smaller, because of how its attention layers cache.

Model comparison

Hermes-4-405B 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 250.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Hermes-4-405B	DeepSeek-V4-Flash
Parameters	406B	291B
Architecture	llama	deepseek4
Layers	126	43
Native context	131,072	1,048,576
Mixture of experts	no	yes, 256 experts
Quantizations published	16	12
Smallest quantization	81.23 GiB	76.87 GiB
Q4_K_M	226.38 GiB	—
Licence	llama3	mit

KV cache by context

the term that decides long-context viability

Context	Hermes-4-405B	DeepSeek-V4-Flash	Ratio
4,096	1.97 GiB	0.06 GiB	31.26×
8,192	3.94 GiB	0.06 GiB	62.51×
16,384	7.88 GiB	0.06 GiB	125.02×
32,768	15.75 GiB	0.06 GiB	250.05×
65,536	31.50 GiB	0.06 GiB	500.09×
131,072	63.00 GiB	0.06 GiB	1000.19×