Model comparison

Ministral-3-3B-Instruct-2512-BF16 vs Qwen3-VL-8B-Instruct-abliterated-v1

At Q4_K_M, Ministral-3-3B-Instruct-2512-BF16 is the smaller download — 2,146,495,744 bytes against 5,027,785,792.

From the file· summed bytes, KV per layer

Side by side

Ministral-3-3B-Instruct-2512-BF16Qwen3-VL-8B-Instruct-abliterated-v1
Parameters4.3B8.8B
Architecturemistral3qwen3vl
Layers26
Native context262,144
Mixture of expertsnono
Quantizations published3748
Smallest quantization1.23 GiB1.97 GiB
Q4_K_M2.00 GiB4.68 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextMinistral-3-3B-Instruct-2512-BF16Qwen3-VL-8B-Instruct-abliterated-v1Ratio
4,0960.41 GiB
8,1920.81 GiB
16,3841.63 GiB
32,7683.25 GiB
65,5366.50 GiB
131,07213.00 GiB