Last updated: October 11, 2026

Qwen4 GGUF: where to download it, and which file to pick

There is no official Qwen4 GGUF yet, because Qwen4 weights have not been released. When Qwen4-27B ships, GGUF builds for llama.cpp, Ollama and LM Studio usually follow within a day. Here is where to look and how to choose.

Short version: wait for the official release, then download Q4_K_M from the Qwen organization or a well-known quantizer. Ignore “Qwen4” GGUF files that appear before Alibaba publishes the weights.

Where Qwen4 GGUF files will appear

Qwen on Hugging Face. Check first. Qwen sometimes publishes its own GGUF builds; the base model must be a Qwen/ repository.
Unsloth. Published Qwen3.8-27B GGUF (unsloth/Qwen3.8-27B-GGUF), the most downloaded build of that model.
LM Studio community. Published lmstudio-community/Qwen3.8-27B-GGUF, tuned for LM Studio.
Ollama library. Lists Qwen3.8 as qwen3.8:27b (about 18 GB). Expect a qwen4 entry after release.

Source: Hugging Face search: qwen4 gguf

Source: unsloth/Qwen3.8-27B-GGUF

Source: lmstudio-community/Qwen3.8-27B-GGUF

Source: Ollama library: qwen3.8

Which quantization to pick

Format	Weights	Total, 8K context	Total, 32K	Total, 128K
BF16	54 GB	56 GB	57 GB	64 GB
Q8_0	29 GB	30 GB	32 GB	38 GB
Q6_K	22 GB	24 GB	25 GB	32 GB
Q5_K_M	19 GB	21 GB	22 GB	29 GB
Q4_K_M	17 GB	18 GB	20 GB	26 GB
IQ4_XS	14 GB	16 GB	18 GB	24 GB
Q3_K_M	13 GB	15 GB	16 GB	23 GB

Estimates for a 27B model with the Qwen3.8-27B layout, including KV cache and about 1 GB of runtime overhead. Q4_K_M is the usual balance of size and quality; go to Q5_K_M or Q6_K if memory allows.

Calculate for your own GPU and context

How to avoid fake or broken files

No official weights means no genuine Qwen4 GGUF. Anything older than the release is a rename or a fine-tune of another model.
Check the model card’s base model. It should point to the official Qwen/ repository.
Skip “uncensored”, “abliterated” or merged builds if you want the original model.
If Qwen4 uses the new Qwen3.8-Flash-Next architecture, older llama.cpp builds may refuse to load it. Update llama.cpp, Ollama or LM Studio first.

Source: Qwen3.8-Flash-Next model card (Qwen4 architecture preview)

Run the Qwen3.8-27B GGUF today

Use this to test your machine now. When Qwen4-27B GGUF files appear, change the repository name.

# llama.cpp (downloads from Hugging Face)
llama-server -hf unsloth/Qwen3.8-27B-GGUF:Q4_K_M -c 32768

# Ollama
ollama run qwen3.8:27b

Source: unsloth/Qwen3.8-27B-GGUF

Source: Ollama library: qwen3.8

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

FAQ

Is there a Qwen4 GGUF?

Not an official one. Qwen4 weights are not released as of October 11, 2026, so no genuine Qwen4 GGUF exists yet.

Which Qwen4 GGUF should I download?

Q4_K_M for most 24 GB GPUs, Q5_K_M or Q6_K if you have 32 GB, Q8_0 for 48 GB. Use the VRAM calculator to check your context length.

Will Qwen4 GGUF work in Ollama and LM Studio?

Once their llama.cpp version supports the Qwen4 architecture. Update the app before you download.

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.