Model comparison

MiniMax-H3 vs Wan-Dancer-14B

At Q4_K_M, MiniMax-H3 is the smaller download — 14,576,977,960 bytes against 26,083,251,744.

From the file· summed bytes, KV per layer

Side by side

MiniMax-H3Wan-Dancer-14B
Parameters33.1B17.2B
Architectureqwen3vlwan
Layers
Native context
Mixture of expertsnono
Quantizations published225
Smallest quantization10.60 GiB15.83 GiB
Q4_K_M13.58 GiB24.29 GiB
Licenceotherapache-2.0

KV cache by context

the term that decides long-context viability
ContextMiniMax-H3Wan-Dancer-14BRatio
4,096
8,192
16,384
32,768
65,536
131,072