About
V-JEPA / JEPA is a family of AI models developed by the FAIR team at Meta, built on the Joint-Embedding Predictive Architecture (JEPA) proposed in Yann LeCun's vision for advanced machine intelligence (AMI). JEPA models learn in a self-supervised way by predicting the representations (embeddings) of missing parts of the data in latent space, rather than reconstructing raw pixels. The first video model, V-JEPA, was released in 2024. In 2025 Meta introduced V-JEPA 2 — a world model trained on over one million hours of video, together with an action-conditioned variant (V-JEPA 2-AC) for robotic task planning. The family is developed as a foundation for world models capable of understanding, prediction and planning in the physical environment.