Tier 2 Restricted open-weight
What the publisher claims: open source (repo)
The one-liner: MIT code around weights nobody ever licensed.
The six questions
Weights mirrored widely and downloadable.
Code is MIT but the weights carry no license grant; the model card disclaims deployed use.
Runs anywhere PyTorch runs.
Corpus never released; filtering described only at a high level.
Not reproducible without the 400M-pair corpus.
OpenAI's CLIP repository merged a large number of outside pull requests to its training and evaluation code, but the model and its unlicensed weights were developed in closed and no outside contributor has a say in direction.
License
MIT (code); model card carries no license tag for the weights — license source ↗
- Model card: 'out of scope' for all deployed use cases; no explicit weight license grant
- Training corpus (WebImageText, 400M pairs) never released
What is actually released
- code (MIT)
- CLIP weights on HF (no license tag)
Why it matters
The historical precedent for the whole weights-without-license pattern that later open-weight releases formalized.
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.