Model comparison

MiniMax-H3 vs Wan_2.1_ComfyUI_repackaged

At Q4_K_M, MiniMax-H3 is the smaller download — 14,576,977,960 bytes against 68,351,109,664.

From the file· summed bytes, KV per layer

Side by side

MiniMax-H3Wan_2.1_ComfyUI_repackaged
Parameters33.1B
Architectureqwen3vlpig
Layers
Native context
Mixture of expertsnono
Quantizations published2248
Smallest quantization10.60 GiB4.25 GiB
Q4_K_M13.58 GiB63.66 GiB
Licenceotherapache-2.0

KV cache by context

the term that decides long-context viability
ContextMiniMax-H3Wan_2.1_ComfyUI_repackagedRatio
4,096
8,192
16,384
32,768
65,536
131,072