Tier 3 Practical open-weight
What the publisher claims: open source, open weights
The one-liner: Fully permissive self-supervised vision backbone.
The six questions
Weights and code downloadable, ungated.
Apache 2.0 across code and released weights; only specialized variants are NC.
Standard PyTorch.
LVD-142M curation described in the paper; dataset itself not released.
Curating an equivalent LVD-scale dataset is not practically reproducible.
Meta's DINOv2 code and Apache-2.0 weights are public and the research repo takes the occasional outside PR, but there is no community roadmap and Meta alone controls what the next backbone looks like.
License
Apache 2.0 (code and released weights) OSI-approved — license source ↗
- X-ray DINO weights: FAIR Noncommercial Research License
- Cell-DINO code: CC-BY-NC
What is actually released
- training/eval code
- ViT weights (giant down to small, Apache-2.0 tagged)
- DINOv2 evaluation suite
Why it matters
The earlier CC-BY-NC framing aged out — DINOv2 shows a lab re-releasing its vision line under a true open-source license.
Sources
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.