Model comparison

AREX-Turbo vs Qwen3.5-9B

At Q4_K_M, AREX-Turbo is the smaller download — 2,884,850,752 bytes against 5,627,044,256.

From the file· summed bytes, KV per layer

Side by side

AREX-TurboQwen3.5-9B
Parameters4.5B9.7B
Architectureqwen35qwen35
Layers3232
Native context262,144262,144
Mixture of expertsnono
Quantizations published2334
Smallest quantization1.70 GiB2.97 GiB
Q4_K_M2.69 GiB5.24 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextAREX-TurboQwen3.5-9BRatio
4,0960.13 GiB0.13 GiB1.00×
8,1920.25 GiB0.25 GiB1.00×
16,3840.50 GiB0.50 GiB1.00×
32,7681.00 GiB1.00 GiB1.00×
65,5362.00 GiB2.00 GiB1.00×
131,0724.00 GiB4.00 GiB1.00×