Tier 1 Source-available or access-limited
What the publisher claims: open weights
The one-liner: Gated weights under a bespoke agreement — open in marketing, licensed in substance.
The six questions
Checkpoints downloadable, but gated behind manual approval.
Royalty-free use with attribution, non-transferable, unilateral amendment rights, and no clear commercial-use grant.
Standard PyTorch once granted.
Two named datasets with sizes documented; data itself not released.
Web-scale curation not reproducible.
DINOv3's gated checkpoints and custom licence are published alongside an open repository that accepts outside pull requests on code only, while the model, its data, and the release decision remain Meta's.
License
Custom DINOv3 License — license source ↗
- Gated HF access with manual approval and contact-info sharing
- Royalty-free but non-transferable grant; redistribution only under the same agreement
- Mandatory 'Built with DINOv3' attribution
- Trade-control and military-use exclusions; Meta may amend terms unilaterally
- No commercial-use clause found in the license text
What is actually released
- 12 checkpoints (10 on LVD-1689M web data, 2 on SAT-493M satellite data)
- fine-tuning notebooks
- eval suite
Why it matters
The clearest vision-side example of openness retreating between generations: DINOv2 Apache, DINOv3 gated custom terms.
Sources
- https://raw.githubusercontent.com/facebookresearch/dinov3/main/LICENSE.md
- https://ai.meta.com/resources/models-and-libraries/dinov3-license
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.