Our verdict
The cleanest licence position among the large open-weight model families, and a useful control when judging the others. Apache-2.0 across the open-weight range, with an express patent grant, no territorial carve-out and no user-count ceiling - the same position Alibaba takes with its Wan video models. The caveats are practical rather than legal: serious hardware, no support, and a hosted Qwen Chat product sitting under entirely separate terms.
Key features
- Open-weight language models published for download
- Apache-2.0 across the open-weight range, per the project's own statement
- Multiple model sizes for different hardware budgets
- Weights distributed through Hugging Face
- Runs through standard local tooling such as Ollama and LM Studio
Strengths
- Apache-2.0 across the open-weight models, stated plainly in the project's own README
- No territorial restriction, no monthly-active-user ceiling and no use-based restrictions attached to the grant
- An express patent grant, which Apache-2.0 carries and MIT does not
- Weights published openly rather than behind a request form
- The same licence position Alibaba takes with its Wan video models, which makes the two easy to reason about together
Limitations
- Running the larger models well needs serious GPU capacity, which is the real cost
- No vendor support or SLA stands behind the open weights
- The README points to the Hugging Face repositories for the licence files rather than shipping them all in one place
- Hosted Qwen Chat is a separate product under separate terms, not covered by Apache-2.0
- Model quality and suitability vary by size, so a clean licence says nothing about fit
Who should skip Qwen
Teams without the hardware to run open weights, who will end up on a hosted API under different terms anyway. Also anyone needing vendor support behind a production language model.
Pricing details
What the free plan includes
- The open weights are free and published, with no allowance to exhaust; hardware is the only constraint. Alibaba also runs a hosted Qwen Chat product, which is a separate offering from the open weights documented here.
Worth knowing
The open-weight models have no price and no paid tier; they are published for download under Apache-2.0, so the cost is hardware and electricity. Alibaba separately offers hosted access, whose pricing we were not able to verify at source for this check and is therefore not published here. Checked 23 October 2026.
Commercial use
Allowed. The project's README states: "All our open-weight models are licensed under Apache 2.0. You can find the license files in the respective Hugging Face repositories." Apache-2.0 permits commercial use, modification, redistribution and running the models as a hosted service for clients, and it carries an express patent grant, requiring only that notices and the licence be retained; it grants no trademark rights. There is no territorial carve-out, no monthly-active-user threshold and no use-based restriction attached to the grant, which is what distinguishes it from several other open-weight families we have read. One scoping note: this covers the open-weight models only. Alibaba's hosted Qwen Chat service is a separate product governed by its own terms rather than by Apache-2.0, and those terms are not covered here. Checked 23 October 2026.
Notable integrations
- Ollama and LM Studio for local running
- Hugging Face for weight distribution
- Standard inference stacks that accept open weights
- A separate hosted Qwen Chat service from Alibaba
Alternatives to Qwen
-
Wan
Alibaba's open-weight video generation model under Apache-2.0, with the project stating plainly that it claims no rights over what you generate.
-
Ollama
A local model runtime. It downloads, manages and serves open-weight language models on your own machine, and exposes them over an HTTP API on localhost.
-
LM Studio
A desktop app that is both model runtime and chat interface. It finds and downloads models, runs them locally, and can serve them to other software over an OpenAI-compatible API.