Model comparison

gemma-3-12b-it-vl-GLM-4.7-Flash-Heretic-Uncensored-Thinking vs Qwen3-8B

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: gemma-3-12b-it-vl-GLM-4.7-Flash-Heretic-Uncensored-Thinking's KV cache at 32K is 1.8× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

gemma-3-12b-it-vl-GLM-4.7-Flash-Heretic-Uncensored-ThinkingQwen3-8B
Parameters12.2B8.2B
Architecturegemma3qwen3
Layers4836
Native context131,07240,960
Mixture of expertsnono
Quantizations published2451
Smallest quantization2.74 GiB2.12 GiB
Q4_K_M—4.68 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
Contextgemma-3-12b-it-vl-GLM-4.7-Flash-Heretic-Uncensored-ThinkingQwen3-8BRatio
4,0960.72 GiB0.56 GiB1.28×
8,1920.97 GiB1.13 GiB1.16×
16,3841.47 GiB2.25 GiB1.53×
32,7682.47 GiB4.50 GiB1.82×
65,5364.47 GiB9.00 GiB2.01×
131,0728.47 GiB18.00 GiB2.13×
gemma-3-12b-it-vl-GLM-4.7-Flash-Heretic-Uncensored-Thinking vs Qwen3-8B — size, memory and hardware fit — ossmodeldb