Qwen 3 Coder snap

The Qwen 3 Coder snap is published by Canonical (Ubuntu publisher on the Snap Store) and installs an optimized environment for running Alibaba's Qwen3 Coder model locally. It ships as a snap for amd64 and arm64 and is launched from the command line with qwen3-coder. The latest/stable channel lists version 2.0.0-beta.14+7e755bd, with a newer 2.0.0-beta.17+66f4d4c on the beta and edge channels.

What It Is

This is a packaged local-inference runtime for Qwen3 Coder, which the listing describes as an instruction-tuned language model from Alibaba specialized in coding that can switch from deep reasoning to fast response. Instead of assembling inference dependencies by hand, users install one snap and run the model on their own hardware.

Hardware-Optimized Engines

The snap bundles three inference engines: a CPU engine optimized for x64 and ARM (armv8, armv9) processors, an AMD GPU engine with ROCm acceleration, and an NVIDIA GPU engine with CUDA acceleration. The most suitable engine is selected automatically based on the available hardware. Some accelerators need extra drivers, which are covered in the inference snaps documentation.

Setup Path

Install with sudo snap install qwen3-coder, then run qwen3-coder. The repository can also be built from source by cloning it with submodules and using the make targets make init, make build and make install. Documentation, discussions and the issue tracker are hosted under the Canonical inference-snaps project.

Licensing

The repository is GPL-3.0 licensed. The Qwen3 Coder model itself is provided by Alibaba under the Apache License 2.0, and licenses for bundled software are placed under /snap/qwen3-coder/current/usr/share/doc after installation.