Model comparison

UI-TARS-72B-DPO vs gemma-4-26B-A4B-it-qat-q4_0-unquantized

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: gemma-4-26B-A4B-it-qat-q4_0-unquantized's KV cache at 32K is 6.5× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

UI-TARS-72B-DPOgemma-4-26B-A4B-it-qat-q4_0-unquantized
Parameters73.4B26.5B
Architectureqwen2vlgemma4
Layers8030
Native context32,768262,144
Mixture of expertsnoyes, 128 experts
Quantizations published261
Smallest quantization22.11 GiB13.45 GiB
Q4_K_M44.16 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextUI-TARS-72B-DPOgemma-4-26B-A4B-it-qat-q4_0-unquantizedRatio
4,0961.25 GiB0.45 GiB2.78×
8,1922.50 GiB0.61 GiB4.13×
16,3845.00 GiB0.92 GiB5.45×
32,76810.00 GiB1.54 GiB6.48×
65,53620.00 GiB2.79 GiB7.16×
131,07240.00 GiB5.29 GiB7.56×