Model comparison

gemma-4-26B-A4B-it vs pixtral-12b

At Q4_K_M, pixtral-12b is the smaller download — 7,477,204,608 bytes against 16,796,017,248. At long context the gap widens: gemma-4-26B-A4B-it's KV cache at 32K is 3.2× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

gemma-4-26B-A4B-itpixtral-12b
Parameters26.5B12.7B
Architecturegemma4llama
Layers3040
Native context262,1441,024,000
Mixture of expertsyes, 128 expertsno
Quantizations published4632
Smallest quantization8.99 GiB3.85 GiB
Q4_K_M15.64 GiB6.96 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
Contextgemma-4-26B-A4B-itpixtral-12bRatio
4,0960.45 GiB0.63 GiB1.39×
8,1920.61 GiB1.25 GiB2.06×
16,3840.92 GiB2.50 GiB2.72×
32,7681.54 GiB5.00 GiB3.24×
65,5362.79 GiB10.00 GiB3.58×
131,0725.29 GiB20.00 GiB3.78×