Model comparison

Qwen3-VL-4B-Thinking vs Qwen3-VL-4B-Instruct

At Q4_K_M, Qwen3-VL-4B-Thinking is the smaller download — 2,497,281,472 bytes against 2,497,281,568.

From the file· summed bytes, KV per layer

Side by side

Qwen3-VL-4B-ThinkingQwen3-VL-4B-Instruct
Parameters4.4B4.4B
Architectureqwen3vlqwen3vl
Layers3636
Native context262,144262,144
Mixture of expertsnono
Quantizations published4326
Smallest quantization1.01 GiB1.01 GiB
Q4_K_M2.33 GiB2.33 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextQwen3-VL-4B-ThinkingQwen3-VL-4B-InstructRatio
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×