Qwen3-Next 80B-A3B locally: an 80B at home, if you can hold it

Qwen3-Next 80B-A3B activates only 3B parameters per token — so it's fast. But 80B of weights still have to live somewhere. The honest VRAM math, the…

Aliteq
Lena Fischer · AI & Local Compute Editor

The short answer

Qwen3-Next 80B-A3B is a mixture-of-experts model: 80B parameters in total but only about 3B active per token, so it generates quickly. The catch is storage — at Q4_K_M the full weights need about 47…

80B total / 3B active (MoE, 512 experts, 10 per token), 256K context, Apache-2.0 — from the card.

Q4_K_M ~47.5 GB full weights (computed by our VRAM engine, 16K context) → a 48GB card, or offload.

"3B active" is speed, not memory: all 80B must sit in VRAM or system RAM.

Aliteq

Read the full story

Qwen3-Next 80B-A3B locally: an 80B at home, if you can hold it

Read the full story on Aliteq