Last updated: October 11, 2026

Run Qwen4 locally

Qwen4-27B is the Qwen4 tier meant for local deployment, but its weights are not out yet. Set up your runtime now with Qwen3.8-27B; when Qwen4-27B ships, only the model name changes.

Short version: Ollama on a 24 GB GPU or a Mac with 32 GB or more is the easiest path. A 4-bit build with a 32K context needs about 20 GB.

Pick a runtime

Ollama. One command, runs on Windows, macOS and Linux. Best first choice.
LM Studio. Desktop app with a model browser and a chat window; no terminal needed.
llama.cpp. Most control over quantization, context length and GPU offload.
vLLM or SGLang. For serving an OpenAI-compatible API to many users on NVIDIA GPUs; needs the full or FP8 weights.

Steps

Check your memory with the VRAM calculator and choose a quantization.
Install and update your runtime, so it supports the newest Qwen architecture.
Run Qwen3.8-27B now with one of the commands below to confirm everything works.
When Qwen4-27B is released, replace the model name with the official one from the Qwen Hugging Face page.

VRAM calculator · GGUF downloads

Commands that work today (Qwen3.8-27B)

# Ollama
ollama run qwen3.8:27b

# llama.cpp
llama-server -hf unsloth/Qwen3.8-27B-GGUF:Q4_K_M -c 32768

# vLLM
pip install vllm
vllm serve Qwen/Qwen3.8-27B

# SGLang
pip install sglang
python3 -m sglang.launch_server --model-path Qwen/Qwen3.8-27B --port 30000

The official Qwen4-27B repository name is not published yet. Do not guess it: use the name on the Qwen Hugging Face page when it appears.

Source: Qwen3.8-27B model card (vLLM and SGLang commands)

Source: Ollama library: qwen3.8

On a Mac

Apple Silicon Macs share memory between CPU and GPU, but macOS lets the GPU use only part of it by default. For a 4-bit 27B model with a 32K context (about 20 GB), 32 GB of memory is the realistic minimum and 48 GB or more is comfortable. Ollama also lists an MLX build of Qwen3.8 (qwen3.8:27b-mlx).

Source: Ollama library: qwen3.8

Qwen4 status and release date · Qwen4-27B · VRAM calculator · GGUF downloads · Try online · Qwen4-27B vs Qwen3.8-27B

FAQ

Can I run Qwen4 locally now?

Not Qwen4 itself: no weights are released as of October 11, 2026. You can run Qwen3.8-27B with the same tools today.

What is the easiest way to run Qwen4-27B?

Ollama. Once Qwen4 is in its library, it will be one command, as qwen3.8:27b is now.

Do I need an NVIDIA GPU?

No. Ollama, LM Studio and llama.cpp also run on Apple Silicon Macs and AMD GPUs. vLLM and SGLang are mainly for NVIDIA servers.

Independent guide, not affiliated with Alibaba or the Qwen team. Status checked on October 11, 2026. Qwen4 weights are not released yet; we will update this page when they are.