Qwen3-Coder-30B-A3B-Instruct is a Mixture-of-Experts (MoE) code-focused instruction model developed by Qwen Team, Alibaba Cloud. It features 30B total parameters with ~3B active parameters per token.

For full details, see the original model page.

Usage

llama.cpp

bash

llama-cli \
  -m Qwen3-Coder-30B-A3B-Instruct-f16-Q4_K_M.gguf \
  --chat-template qwen3 \
  -p "Write a Python function that sorts a list of dictionaries by a given key." \
  -n 512

llama-server

bash

llama-server \
  -m Qwen3-Coder-30B-A3B-Instruct-f16-Q4_K_M.gguf \
  --chat-template qwen3 \
  --port 8080

Ollama (via Modelfile)

FROM ./Qwen3-Coder-30B-A3B-Instruct-f16-Q4_K_M.gguf
PARAMETER num_ctx 32768
TEMPLATE "{{ ... }}"  # use Qwen3 chat template

Quantization details

File	Quant	Size (approx.)
`Qwen3-Coder-30B-A3B-Instruct-f16-Q4_K_M.gguf`	Q4_K_M	~17 GB

Q4_K_M uses 4-bit quantization with K-quant method on most layers, providing a good balance between size and quality.

License

This quantized model is derived from Qwen/Qwen3-Coder-30B-A3B-Instruct and is released under the same Apache 2.0 License.

Per Qwen's terms, appropriate credit is given to the original authors:

Qwen3-Coder-30B-A3B-Instruct is developed by Qwen Team, Alibaba Cloud. Original model: https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct

Citation

bibtex

@misc{qwen3coder,
  title        = {Qwen3-Coder},
  author       = {Qwen Team},
  year         = {2025},
  organization = {Alibaba Cloud},
  url          = {https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct}
}

Zoed/Qwen3-Coder-30B-A3B-Instruct

Qwen3-Coder-30B-A3B-Instruct · Q4_K_M GGUF

About the base model