Qwen's new image model leaves Apache 2.0 for a research-only licence
EditorialBy TrustList Editorial
Every earlier Qwen-Image release was Apache 2.0. Qwen-Image-2.1, published on 20 September, is licensed for research and evaluation only, under Chinese law, with commercial use needing a separate licence. The announcement calls it open-source.
About Qwen's new image model leaves Apache 2.0 for a research-only licence
Qwen's new image model leaves Apache 2.0 for a research-only licence
20 September 2026 — Alibaba's Qwen team published Qwen-Image-2.1 on 20 September: a 7-billion-parameter image generation and editing model with native transparent output and editing from up to ten reference images. For a business, the most important thing about it is not in the announcement at all. It is the licence.
What changed
We read the licence tag on every Qwen-Image model repository the vendor publishes on its model host. The six earlier models — the original release in August 2025, the edit models of August and September 2025, and the December 2025 updates, including the layered-transparency model — are all published under Apache 2.0, a permissive licence that allows commercial use.
Qwen-Image-2.1 is published under the Qwen Research Licence Agreement, dated the same day. It grants the right to use, modify and redistribute the model for non-commercial purposes only, and defines non-commercial as "for research or evaluation purposes only". Commercial use requires a separate licence, requested from the vendor by email. The agreement is governed by the laws of China, gives the courts of Hangzhou exclusive jurisdiction, requires a "Built with Qwen" or "Improved using Qwen" notice on any model trained or improved with the materials or their outputs, and terminates for anyone who brings an intellectual-property claim against the licensor over the materials.
The vendor's announcement describes the release as open-source and does not mention the licence.
Why it matters
Open-weight model families are usually adopted on the assumption that the next version will come under the same terms as the last. This release breaks that assumption in the direction that matters. A team that built a product on the Apache-licensed Qwen-Image models can keep using those versions on the terms they were published under, since an Apache 2.0 grant is perpetual; what it cannot do is upgrade to 2.1 in a commercial product without a separate agreement.
The risk is less that anyone will misread the licence than that nobody will read it. Model upgrades are often a one-line change in a configuration file, made by an engineer testing whether the new version looks better — which, by the vendor's account, it may.
What to do
Check your model registry, container images and configuration for any reference to Qwen-Image-2.1 in a commercial system, and pin the Apache-licensed versions where they are in use. If the new version is wanted, request the commercial licence before shipping it and read its governing-law and venue terms with whoever signs your contracts. And add the licence to the checks your team runs on every model version, not only on every new model: this release shows the terms can change between point releases of the same family.
Sources
- Qwen Research Licence Agreement for Qwen-Image-2.1, model host (release date 20 September 2026) — 20 September 2026
- Qwen blog, Qwen-Image-2.1 announcement (read 21 September 2026) — 21 September 2026
- Qwen-Image, the original release, model card with Apache 2.0 licence (read 21 September 2026) — 21 September 2026
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ProductsSoftware and SaaS with pricing, features, integrations and alternatives.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- AI ModelsBenchmark scores and community ratings for every major model.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.