Model comparison

MiniMax-H3 vs lingbot-world-v2-14b-causal-fast

At Q4_K_M, lingbot-world-v2-14b-causal-fast is the smaller download — 12,054,188,608 bytes against 14,576,977,960.

From the file· summed bytes, KV per layer

Side by side

MiniMax-H3lingbot-world-v2-14b-causal-fast
Parameters33.1B18.5B
Architectureqwen3vlwan
Layers
Native context
Mixture of expertsnono
Quantizations published225
Smallest quantization10.60 GiB11.23 GiB
Q4_K_M13.58 GiB11.23 GiB
Licenceothercc-by-nc-sa-4.0

KV cache by context

the term that decides long-context viability
ContextMiniMax-H3lingbot-world-v2-14b-causal-fastRatio
4,096
8,192
16,384
32,768
65,536
131,072