Tier 4 Deeply open
What the publisher claims: open source, open weights
The one-liner: The rare vision release where data, code, and weights all shipped.
The six questions
All checkpoints and code downloadable.
Apache 2.0 including checkpoints; dataset is the exception, not the model.
Runs on commodity GPUs.
SA-1B composition documented and the dataset is actually published (under its own terms); training pipeline described.
Dataset release plus code makes a reproduction attempt plausible at cost.
Meta's SAM and SAM 2 repositories merge outside pull requests on code and ship the SA-1B dataset, but model development, checkpoints, and release timing are decided by Meta alone.
License
Apache 2.0 (code, checkpoints, SAM 2.1 included) OSI-approved — license source ↗
- SA-1B dataset under a separate Research License (non-commercial)
What is actually released
- model code
- SAM, SAM 2, SAM 2.1 checkpoints
- demo and training code
- SA-1B dataset (research license)
Why it matters
With SA-1B published, SAM is the vision line that comes closest to end-to-end openness — held back from Tier 5 only by the dataset's separate research license.
Sources
- https://github.com/facebookresearch/segment-anything
- https://github.com/facebookresearch/sam2
- https://huggingface.co/facebook/sam2.1-hiera-large
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.