Model comparison

cogito-v2-preview-deepseek-671B-MoE vs DeepSeek-V4-Flash-0731

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: DeepSeek-V4-Flash-0731's KV cache at 32K is 34.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

cogito-v2-preview-deepseek-671B-MoEDeepSeek-V4-Flash-0731
Parameters671B304B
Architecturedeepseek2deepseek4
Layers6143
Native context163,8401,048,576
Mixture of expertsyes, 256 expertsyes, 256 experts
Quantizations published2011
Smallest quantization150.73 GiB76.87 GiB
Q4_K_M377.13 GiB
Licencemitmit

KV cache by context

the term that decides long-context viability
Contextcogito-v2-preview-deepseek-671B-MoEDeepSeek-V4-Flash-0731Ratio
4,0960.27 GiB0.06 GiB4.26×
8,1920.54 GiB0.06 GiB8.51×
16,3841.07 GiB0.06 GiB17.02×
32,7682.14 GiB0.06 GiB34.05×
65,5364.29 GiB0.06 GiB68.09×
131,0728.58 GiB0.06 GiB136.19×
cogito-v2-preview-deepseek-671B-MoE vs DeepSeek-V4-Flash-0731 — size, memory and hardware fit — ossmodeldb