Model comparison

Qwen3-4B-Instruct-2507-heretic vs Qwen3-8B

At Q4_K_M, Qwen3-4B-Instruct-2507-heretic is the smaller download — 2,497,279,424 bytes against 5,027,783,488.

From the file· summed bytes, KV per layer

Side by side

Qwen3-4B-Instruct-2507-hereticQwen3-8B
Parameters4.0B8.2B
Architectureqwen3qwen3
Layers3636
Native context262,14440,960
Mixture of expertsnono
Quantizations published2551
Smallest quantization1.41 GiB2.12 GiB
Q4_K_M2.33 GiB4.68 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextQwen3-4B-Instruct-2507-hereticQwen3-8BRatio
4,0960.56 GiB0.56 GiB1.00×
8,1921.13 GiB1.13 GiB1.00×
16,3842.25 GiB2.25 GiB1.00×
32,7684.50 GiB4.50 GiB1.00×
65,5369.00 GiB9.00 GiB1.00×
131,07218.00 GiB18.00 GiB1.00×