Qwen3.6 35B-A3B locally: a 35B that thinks like a 3B

Qwen3.6's default is a mixture-of-experts model: 35B on disk, 3B active per token. That makes it fast — but the VRAM math has a twist worth…

Aliteq
Lena Fischer · AI & Local Compute Editor

The short answer

Qwen3.6 35B-A3B is the mixture-of-experts model the bare ollama run qwen3.6 pulls: 35B parameters total, but only ~3B active per token, so it runs fast. Its full weights need about 21 GB at Q4_K_M…

35B total / 3B activated (MoE), 256K context, Apache-2.0 — from the model card.

The bare ollama run qwen3.6 pulls this 35B-A3B (verified on ollama.com/library).

Q4_K_M ~21.4 GB full weights (computed) → a 24GB card; or offload experts to RAM to fit less.

Aliteq

Read the full story

Qwen3.6 35B-A3B locally: a 35B that thinks like a 3B

Read the full story on Aliteq