Instructions to use Ohjaaja/Qwen3-Coder-Next-Q3mix with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Ohjaaja/Qwen3-Coder-Next-Q3mix with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="Ohjaaja/Qwen3-Coder-Next-Q3mix",
	filename="Qwen3-Coder-Next-Q3mix.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use Ohjaaja/Qwen3-Coder-Next-Q3mix with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Ohjaaja/Qwen3-Coder-Next-Q3mix
# Run inference directly in the terminal:
llama-cli -hf Ohjaaja/Qwen3-Coder-Next-Q3mix

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Ohjaaja/Qwen3-Coder-Next-Q3mix
# Run inference directly in the terminal:
llama-cli -hf Ohjaaja/Qwen3-Coder-Next-Q3mix

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf Ohjaaja/Qwen3-Coder-Next-Q3mix
# Run inference directly in the terminal:
./llama-cli -hf Ohjaaja/Qwen3-Coder-Next-Q3mix

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf Ohjaaja/Qwen3-Coder-Next-Q3mix
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Ohjaaja/Qwen3-Coder-Next-Q3mix

Use Docker

docker model run hf.co/Ohjaaja/Qwen3-Coder-Next-Q3mix

LM Studio
Jan
Ollama
How to use Ohjaaja/Qwen3-Coder-Next-Q3mix with Ollama:
```
ollama run hf.co/Ohjaaja/Qwen3-Coder-Next-Q3mix
```

Unsloth Studio new

How to use Ohjaaja/Qwen3-Coder-Next-Q3mix with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Ohjaaja/Qwen3-Coder-Next-Q3mix to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Ohjaaja/Qwen3-Coder-Next-Q3mix to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Ohjaaja/Qwen3-Coder-Next-Q3mix to start chatting

Pi new

How to use Ohjaaja/Qwen3-Coder-Next-Q3mix with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf Ohjaaja/Qwen3-Coder-Next-Q3mix

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Qwen3-Coder-Next-Q3mix"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Ohjaaja/Qwen3-Coder-Next-Q3mix with Docker Model Runner:
```
docker model run hf.co/Ohjaaja/Qwen3-Coder-Next-Q3mix
```

Lemonade

How to use Ohjaaja/Qwen3-Coder-Next-Q3mix with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Ohjaaja/Qwen3-Coder-Next-Q3mix

Run and chat with the model

lemonade run user.Qwen3-Coder-Next-Q3mix-{{QUANT_TAG}}

List all available models

lemonade list

Base model: Qwen/Qwen3-Coder-Next

Perplexity (WikiText-2 test set, ctx=512, 584 chunks)

Model	PPL	±
Qwen3-Coder-Next-Q3mix (this)	8.4837	0.06573
Ubergarm smol IQ3_KS (reference)	8.4649	0.06633

Model is compatible with mainline llama.cpp. However test was ran with llama.cpp TurboQuant build](https://github.com/TheTom/llama-cpp-turboquant) Take note benchmark was not ran using mainline llama.cpp and ubergarm model was benchmarked using ik_llama.cpp so test result is not exactly comparable.

This quant is inspired by ubergarm / Qwen3-Coder-Next-GGUF smol-IQ3_KS 30.728 GiB (3.313 BPW)

Qwen3-Coder-Next-Q3mix was quantized using:

#!/usr/bin/env bash cat > /tmp/qwen3_tensors.txt << 'EOF' attn_gate.weight=q6_k
attn_qkv.weight=q6_k
attn_output.weight=q6_k
attn_q.weight=q6_k
attn_k.weight=q6_k
attn_v.weight=q6_k
ssm_ba.weight=q6_k
ssm_out.weight=q6_k
ffn_down_shexp.weight=q6_k
ffn_gate_shexp.weight=q6_k
ffn_up_shexp.weight=q6_k
ffn_gate_inp.weight=q8_0
ffn_gate_inp_shexp.weight=q8_0
ffn_down_exps.weight=iq3_s
ffn_gate_exps.weight=iq3_s
ffn_up_exps.weight=iq3_s
token_embd.weight=iq4_nl
output.weight=q6_k
EOF

~/Documents/llama-cpp-turboquant/build/bin/llama-quantize
--tensor-type-file /tmp/qwen3_tensors.txt
--imatrix ~/imatrix-Qwen3-Coder-Next-BF16.dat
~/Qwen3-Coder-Next-f16.gguf
~/Qwen3-Coder-Next-Q3mix.gguf
IQ3_S
$(nproc)

So huge thanks for ubergarm for inspiration and expertise. This model is not intended to rival. It's from personal need to run on mainline llama.cpp on 12gb vram and 32gb ram.

Best Practices from the base models card:

To achieve optimal performance, we recommend the following sampling parameters: temperature=1.0, top_p=0.95, top_k=40. pipeline_tag: text-generation

tags:

code

Downloads last month: 285

GGUF

Model size

80B params

Architecture

qwen3next

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support