- What changed
- Researchers introduced WROP, a 1.5M-sample training corpus and exam designed to evaluate object permanence in video models, along with a 16B world model named PWM-WROP.
- Why you should care
- The release provides new data and benchmark resources for studying cognitive capabilities like object permanence in video generation models.
- Your move
- Watch. Monitor adoption and reproducibility of the WROP dataset across other video model architectures.
- What to watch next
- Independent replication of the PWM-WROP evaluation results by other research groups.
- Event
- research
- Event date
- Sep 25, 2026
- Relevant to
- General AI readers
