Tier 3 Practical open-weight
What the publisher claims: open source
The one-liner: The community's answer to CLIP's missing license.
The six questions
Code and many checkpoints downloadable.
Code MIT; specific checkpoints limited by their training-data terms.
Standard PyTorch.
Training datasets are named but themselves carry murky provenance.
Depends on third-party dataset availability.
OpenCLIP is run by a group of independent maintainers across institutions, with public issue-thread roadmaps, routinely merged outside pull requests, and checkpoints added by community contributors, so no single company controls its direction.
License
MIT (code); released weights inherit upstream dataset terms OSI-approved — license source ↗
- Weights trained on LAION and similar datasets carry non-commercial research terms
What is actually released
- code (MIT)
- checkpoints across architectures and datasets
- training scripts
Why it matters
Shows open source re-implementations fixing what the original release left closed — at the cost of inheriting dataset-licensing problems.
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.