Tier 3 Practical open-weight
What the publisher claims: open source, open weights, world model
The one-liner: Permissively licensed video world model — the JEPA line's most open release.
The six questions
Encoder and action-conditioned checkpoints downloadable across the collection.
MIT/Apache-2.0 tags on the released weights; commercial use permitted.
Standard PyTorch; runs on user GPUs.
Method paper and pretraining-corpus description (video) published; corpus and recipe not released.
No path to rebuild the pretrained encoder at scale.
Meta's V-JEPA 2 repository is public and merges outside contributions to evaluation and training utilities, but the world-model research direction and checkpoint releases are decided by FAIR.
License
MIT and Apache-2.0 across released checkpoints (GitHub code split MIT/Apache-2.0) OSI-approved — license source ↗
- Sibling jepa-wms research repo remains CC-BY-NC 4.0
- V-JEPA 2.1 checkpoints distributed via download links, not yet the Meta HF org
What is actually released
- V-JEPA 2 ViT-L/G weights (MIT and Apache-2.0 tags on HF)
- action-conditioned V-JEPA 2-AC checkpoints
- V-JEPA 2.1 checkpoints (download links + torch.hub)
- training/eval code
Why it matters
Contradicts the assumption that Meta's JEPA family is uniformly non-commercial: the world-model generation moved to MIT/Apache while I-JEPA and the research repos stayed CC-BY-NC.
Sources
- https://github.com/facebookresearch/vjepa2
- https://huggingface.co/collection/facebook/vjepa-2
- https://ai.meta.com/research/vjepa
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.