Meta has launched an innovative world model, V-JEPA 2, featuring an impressive 1.2 billion parameters and grounded in the robust JEPA architecture. This cutting-edge model aims to enhance AI's physical reasoning capabilities by meticulously understanding physical realities and accurately predicting outcomes. Through over a million hours of self-supervised learning from diverse videos, V-JEPA 2 has significantly bolstered the AI's proficiencies in comprehension, prediction, and strategic planning.