Model comparison

MiniMax-H3 vs Wan2.1-VACE-14B

At Q4_K_M, Wan2.1-VACE-14B is the smaller download — 11,639,453,600 bytes against 14,576,977,960.

From the file· summed bytes, KV per layer

Side by side

MiniMax-H3Wan2.1-VACE-14B
Parameters33.1B17.3B
Architectureqwen3vlwan
Layers
Native context
Mixture of expertsnono
Quantizations published2213
Smallest quantization10.60 GiB7.31 GiB
Q4_K_M13.58 GiB10.84 GiB
Licenceotherapache-2.0

KV cache by context

the term that decides long-context viability
ContextMiniMax-H3Wan2.1-VACE-14BRatio
4,096
8,192
16,384
32,768
65,536
131,072