V-JEPA 2

V-JEPA 2 is a video-trained model accessible via Web and CLI interfaces that specializes in forecasting physical phenomena and supporting zero-shot robot planning. It provides developers and data scientists with open-access code and benchmarks for advanced visual understanding.

• Supports zero-shot robot planning based on video training data.
• Provides publicly accessible model weights, code, and benchmarks.
• Available across both Web and CLI platforms for flexible implementation.
• Achieves state-of-the-art results in comprehending physical world interactions.

Ideal for: Developers, data scientists, and open source maintainers requiring a video-trained model for AI agents and code development related to physical world modeling.