Model comparison

DeepSeek-V3.1 vs Hy3

At Q4_K_M, Hy3 is the smaller download — 180,999,059,328 bytes against 404,494,158,304. At long context the gap widens: DeepSeek-V3.1's KV cache at 32K is 4.7× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

DeepSeek-V3.1Hy3
Parameters685B299B
Architecturedeepseek2hy_v3
Layers6180
Native context163,840262,144
Mixture of expertsyes, 256 expertsyes, 192 experts
Quantizations published5037
Smallest quantization137.32 GiB59.65 GiB
Q4_K_M376.71 GiB168.57 GiB
Licencemitapache-2.0

KV cache by context

the term that decides long-context viability
ContextDeepSeek-V3.1Hy3Ratio
4,0960.27 GiB1.25 GiB4.66×
8,1920.54 GiB2.50 GiB4.66×
16,3841.07 GiB5.00 GiB4.66×
32,7682.14 GiB10.00 GiB4.66×
65,5364.29 GiB20.00 GiB4.66×
131,0728.58 GiB40.00 GiB4.66×