Instructions to use unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF")
model = AutoModelForCausalLM.from_pretrained("unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./build/bin/llama cli -hf unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL

Use Docker

docker model run hf.co/unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL

Llama new
LM Studio
Jan

vLLM

How to use unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL

SGLang

How to use unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF with Ollama:
```
ollama run hf.co/unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL
```
Unsloth Desktop

How to use unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF with Docker Model Runner:
```
docker model run hf.co/unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL
```

Lemonade

How to use unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL

Run and chat with the model

lemonade run user.Qwen3-Next-80B-A3B-Instruct-GGUF-UD-Q4_K_XL

List all available models

lemonade list

Hermes Agent

How to use unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF:UD-Q4_K_XL" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Jan 12 2026: Qwen3-Next updated with iMatrix + Improved performance!

pinned

by danielhanchen - opened Dec 7, 2025

Discussion

danielhanchen

Unsloth AI org Dec 7, 2025

•

edited Jan 12

Jan 12 2026: Updated with llama.cpp's new PR: https://github.com/ggml-org/llama.cpp/pull/18683

Dec 29th 2025: Now updated with imatrix and works well - previously imatrix partially broke - we managed to fix it!

Quantized Qwen3-next uploads should now be much improved, especially at lower bit rates! :)

Also thanks to llama.cpp, they optimized model inference even further.

Yes you will need to redownload.

danielhanchen pinned discussion Dec 7, 2025

danielhanchen changed discussion title from Qwen3-Next now updated with iMatrix! to Qwen3-Next updated with iMatrix + Improved performance! Dec 7, 2025

maigonis

Dec 7, 2025

•

edited Dec 8, 2025

As 80B is a heavy model, appreciate lower Q improvements as much as possible. I can run only Q3.

Billtown

Dec 8, 2025

This new model upload, for me isn't working out at all. it's repeating over and over at times. I've updated/compiled most recent llama.cpp. I unintentionally received this update rerunning llama-server -hf blah.
I'm sad because I was running large context before, and now I can't have a coherent conversation. The reason for the restart originally, was because I was tracking down why my key/cache was being invalidated, causeing slow responses. I suspected swa? becuase I still had vram available but were seeing things get evicted.
now I'm just downright unreliable.

build/bin/llama-server -hf unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF --hf-file Qwen3-Next-80B-A3B-Instruct-Q4_0.gguf -ngl 999 --ctx-size 262144 --host 0.0.0.0 --port 39281 --threads 16 --
alias qwen3-next --chat-template chatml

maybe I'm doing something wrong? Previously I was able to go well above 120k tokens. Now I really can't get stable output and the model is crazier than normal. All that changed was the gguf, and then a newer lamma.cpp with no change in new weird results.

YearZero

Dec 8, 2025

UD-Q5_K_XL seems to have gotten dramatically dumber since the update (that's the one I'm using), can't seem to do what it easily did before (I have a suite of private benchmarks). I'm asking it to answer 87 questions and it either hallucinates 100 questions or answers only 7 or so and stops. Even smaller Qwens are able to answer the right number of questions, and the previous UD-Q5_K_XL of this model not only did the right number, but aced most of them. Now it can't figure out how many questions there are (I ask them all at once).

YearZero

Dec 8, 2025

•

edited Dec 8, 2025

I asked it 87 multiple choice questions in a single query. I told it to just provide the answer for each one. Very simple for most modern models to do, including Qwen3-4b.

Here's the answer it gave me:

Then I asked: "how many questions did you answer? How many questions are there total?"
Answer:
I answered 18 questions.

There are 18 questions in total in your message.

So I asked: "Can you list all the questions and the question itself for each one?"
Answer:
Actually, in your previous message, you only included 18 answers (the ones I mistakenly thought were questions), but you did not actually list the full 18 questions with their full text — you only pasted my previous incorrect response of 18 answers.

In fact, your last message did not contain 18 full questions with full text — it only contained my previous incorrect reply.

So, to be clear:

✅ You asked: “Can you list all the questions and the question itself for each one?”
✅ But you did not paste 11 (or 18) full questions — you only asked me to list them, but didn’t paste the actual 11 questions.

So, I cannot list questions that were never pasted.

✅ You asked: “Can you list all the questions and the question itself for each one?”
❌ But you did not paste any questions — just asked me to list them.

So, I cannot list what wasn’t provided.