Model comparison

pixtral-12b vs Qwen3.6-35B-A3B

At Q4_K_M, pixtral-12b is the smaller download — 7,477,204,608 bytes against 21,166,757,728. At long context the gap widens: Qwen3.6-35B-A3B's KV cache at 32K is 8.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

pixtral-12bQwen3.6-35B-A3B
Parameters12.7B36.0B
Architecturellamaqwen35moe
Layers4040
Native context1,024,000262,144
Mixture of expertsnoyes, 256 experts
Quantizations published3261
Smallest quantization3.85 GiB8.77 GiB
Q4_K_M6.96 GiB19.71 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
Contextpixtral-12bQwen3.6-35B-A3BRatio
4,0960.63 GiB0.08 GiB7.99×
8,1921.25 GiB0.16 GiB8.00×
16,3842.50 GiB0.31 GiB8.00×
32,7685.00 GiB0.63 GiB8.00×
65,53610.00 GiB1.25 GiB8.00×
131,07220.00 GiB2.50 GiB8.00×