Qwen 3 Coder

Run Qwen 3 Coder locally on Linux to generate coding assistance from an instruction-tuned language model rather than relying on a remote service. Published by Ubuntu, it is delivered as an inference snap that packages Qwen3 Coder for local execution. The model is specialized for programming tasks and can switch between deeper reasoning and quicker responses according to the work at hand.

The inference-snap runtime detects the host machine’s hardware and selects runtime and model-weight optimizations suited to its capabilities. Its unified mechanism can make use of available CPU, GPU, or NPU resources without requiring separate hardware-specific tuning. Users can control the snap through the command line, including the qwen3-coder --help command, and applications can access its local inference service through an OpenAI API.

It is intended for developers, programmers, and application builders who want coding assistance in an IDE, chatbot, or media-editing workflow. Developers can also integrate its standardized local API into software that adds generative AI features, while inference-snap services support running multiple model sizes in parallel.

The current release is 2.0.0-beta.14+7e755bd for amd64 and arm64 systems, and it requires Linux with snapd plus the appropriate host drivers. The product is freeware; the Qwen3 Coder model itself is licensed under Apache 2.0, and licenses for bundled software are included at /snap/qwen3-coder/current/usr/share/doc.