Model comparison

Krea-2-Raw vs Qwen-Image-Edit-2511

At Q4_K_M, Krea-2-Raw is the smaller download — 7,486,289,184 bytes against 11,657,701,984.

From the file· summed bytes, KV per layer

Side by side

Krea-2-RawQwen-Image-Edit-2511
Parameters12.8B20.4B
Architecturekrea2qwen_image
Layers
Native context
Mixture of expertsnono
Quantizations published2936
Smallest quantization2.34 GiB4.13 GiB
Q4_K_M6.97 GiB10.86 GiB
Licenceotherapache-2.0

KV cache by context

the term that decides long-context viability
ContextKrea-2-RawQwen-Image-Edit-2511Ratio
4,096
8,192
16,384
32,768
65,536
131,072