Model comparison

MiniCPM-V-4.6 vs Qwen3-VL-2B-Instruct

At Q4_K_M, MiniCPM-V-4.6 is the smaller download — 529,101,504 bytes against 1,107,409,952. At long context the gap widens: MiniCPM-V-4.6's KV cache at 32K is 9.3× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

MiniCPM-V-4.6Qwen3-VL-2B-Instruct
Parameters1.3B2.1B
Architectureqwen35qwen3vl
Layers2428
Native context262,144262,144
Mixture of expertsnono
Quantizations published1546
Smallest quantization0.47 GiB0.50 GiB
Q4_K_M0.49 GiB1.03 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextMiniCPM-V-4.6Qwen3-VL-2B-InstructRatio
4,0960.05 GiB0.44 GiB9.33×
8,1920.09 GiB0.88 GiB9.33×
16,3840.19 GiB1.75 GiB9.33×
32,7680.38 GiB3.50 GiB9.33×
65,5360.75 GiB7.00 GiB9.33×
131,0721.50 GiB14.00 GiB9.33×
MiniCPM-V-4.6 vs Qwen3-VL-2B-Instruct — size, memory and hardware fit — ossmodeldb