Qwen models you can run locally

Ultra-light Qwen 3 model for constrained devices

Ultra-tiny model for embedded and edge

Tiny open text-to-video — 480p clips on 8GB consumer GPUs

Ultra-lightweight coding model

Compact multilingual Qwen 3

Small multimodal Qwen 3.5

Small multimodal Qwen 3.5

Compact dedicated vision-language model — OCR & image chat on edge

Unified text/image-to-video — the local sweet spot under Apache 2.0

8-step distilled image model — photorealism and bilingual text on 16GB cards

Dedicated coding model

Qwen 3 with thinking mode support

The community-favourite local VLM — superb OCR, receipts & captioning

Multimodal Qwen 3.5 mid-size

Strong all-rounder with thinking mode

Open text-to-image with strong English and Chinese typography

Flagship dense Qwen 3.8 — native multimodal all-rounder with video understanding

Flagship open Wan 2.2 — 14B-active MoE for photoreal text-to-video

Flagship native multimodal Qwen 3.5

Flagship dense Qwen 3.6 — native multimodal all-rounder

MoE with only 3.3B active — extremely efficient

Efficient agentic coding MoE — 3B active, 256K context

Efficient vision MoE — 3B active, strong temporal & document understanding

Qwen 3 flagship dense model

Efficient multimodal MoE with 3B active

Big-model quality at 3B-active speed — the mid-hardware sweet spot

High-sparsity MoE — extreme low activation ratio for fast inference at 80B scale

Ultra-efficient agentic coding MoE optimized for tool-calling coding agents

Large multimodal MoE

Massive MoE with 22B active — frontier quality

Flagship vision-language MoE — frontier multimodal reasoning and agentic GUI control

Largest multimodal Qwen 3.5 MoE

Largest open coding MoE — 35B active

Frontier Qwen 3.8 MoE — 95B active, 1M context