Model comparison

MiniMax-H3 vs JoyAI-Echo

At Q4_K_M, MiniMax-H3 is the smaller download — 14,576,977,960 bytes against 15,694,613,568.

From the file· summed bytes, KV per layer

Side by side

MiniMax-H3JoyAI-Echo
Parameters33.1B12.2B
Architectureqwen3vljoyai-gemma
Layers
Native context
Mixture of expertsnono
Quantizations published225
Smallest quantization10.60 GiB8.54 GiB
Q4_K_M13.58 GiB14.62 GiB
Licenceother

KV cache by context

the term that decides long-context viability
ContextMiniMax-H3JoyAI-EchoRatio
4,096
8,192
16,384
32,768
65,536
131,072