Model comparison

Voxtral-Mini-3B-2507 vs Voxtral-Mini-4B-Realtime-2602

At Q4_K_M, Voxtral-Mini-3B-2507 is the smaller download — 2,473,002,304 bytes against 2,830,493,984.

From the file· summed bytes, KV per layer

Side by side

Voxtral-Mini-3B-2507Voxtral-Mini-4B-Realtime-2602
Parameters4.7B4.4B
Architecturevoxtralvoxtral_realtime
Layers3026
Native context131,072131,072
Mixture of expertsnono
Quantizations published318
Smallest quantization1.45 GiB2.35 GiB
Q4_K_M2.30 GiB2.64 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextVoxtral-Mini-3B-2507Voxtral-Mini-4B-Realtime-2602Ratio
4,0960.47 GiB
8,1920.94 GiB
16,3841.88 GiB
32,7683.75 GiB
65,5367.50 GiB
131,07215.00 GiB
Voxtral-Mini-3B-2507 vs Voxtral-Mini-4B-Realtime-2602 — size, memory and hardware fit — ossmodeldb