Model comparison

Kimi-K2.6 vs Kimi-K2.7-Code

These two publish different quantization sets; the table below has the exact sizes.

From the file· summed bytes, KV per layer

Side by side

Kimi-K2.6Kimi-K2.7-Code
Parameters1059B1059B
Architecturedeepseek2deepseek2
Layers6161
Native context262,144262,144
Mixture of expertsyes, 384 expertsyes, 384 experts
Quantizations published1619
Smallest quantization193.16 GiB283.04 GiB
Q4_K_M
Licenceotherother

KV cache by context

the term that decides long-context viability
ContextKimi-K2.6Kimi-K2.7-CodeRatio
4,0960.27 GiB0.27 GiB1.00×
8,1920.54 GiB0.54 GiB1.00×
16,3841.07 GiB1.07 GiB1.00×
32,7682.14 GiB2.14 GiB1.00×
65,5364.29 GiB4.29 GiB1.00×
131,0728.58 GiB8.58 GiB1.00×