Tier 2 Restricted open-weight
What the publisher claims: open-weight, open-source
The one-liner: A strong Apache-2.0 image model whose successor reclassified itself as research-only and whose newest version has no weights.
The six questions
Qwen-Image and Qwen-Image-Edit weights are downloadable; the 3.0 generation is the exception, not the rule for this entry.
The base generation is Apache-2.0, but 2.1 shifts to a research license with its own conditions, so permission varies by release.
Deployable locally through Diffusers and ComfyUI with published inference code.
Model cards and reports describe architecture, resolution, and evaluation; training data and recipe are not published.
Limited.
No public code repo exists — weights-only on the Hub under a research license.
License
Apache-2.0 for Qwen-Image and Qwen-Image-Edit; Qwen Research License Agreement for Qwen-Image-2.1 — license source ↗
- The 2.1 release (September 2026) is governed by a Qwen Research License Agreement rather than Apache-2.0
- Qwen-Image-3.0 ships no weights at all, so the newest capability is not downloadable
- Permissions must be checked at each release, not at the family level
What is actually released
- weights (base and Edit)
- inference code
- model cards and reports
- image editing and text rendering evaluations
Why it matters
Shows how a permissive open-weight launch can be quietly replaced by a research license and an API-only roadmap — the label survives, the terms change.
Sources
- https://huggingface.co/Qwen/Qwen-Image
- https://cellcog.ai/blog/qwen-image-2-1
- https://rits.shanghai.nyu.edu/ai/qwen-image-3-0-4-5k-token-prompts-no-benchmarks-no-weights
Derived from the openness dataset behind But Is It Open? — snapshot 2026-11-07 ↗. Scores are qualitative readings of the cited evidence, not verdicts; tier measures openness, not intelligence.