BAAI Unveils Physis, First General World Foundation Model

The Beijing Academy of Artificial Intelligence (BAAI) unveiled Physis-v0.1 at its 8th annual conference on June 13, 2026. BAAI describes Physis as the world's first general world foundation model — a system designed not merely to process language but to understand and predict how the physical world behaves. Large language models excel at pattern recognition in text but struggle with physical intuition. Humans instinctively know that a glass will shatter if dropped, that fire is hot, or that a heavy object on a slope will roll downhill. AI systems lack this common-sense understanding. Physis aims to bridge that gap by learning from multimodal inputs — text, images, video, and physical simulation data — to build internal representations of causal relationships and physical laws. Wang Zhongyuan, president of BAAI, emphasized that current AI systems face significant limitations when deployed in real-world environments like robotics. A robot can recognize an object but may not understand that it's fragile or dangerous. World models could transform embodied AI by enabling systems to predict outcomes before acting. The applications extend beyond robotics. Scientific research, digital twins, simulation, and autonomous systems all require AI that can model complex real-world processes. Andrew Barto, a Turing Award laureate who spoke at the conference, suggested that combining deep reinforcement learning with knowledge of the brain's reward systems could drive the next wave of AI progress. This development positions Chinese AI research at the frontier of a new paradigm — moving from language understanding to physical world understanding — which could have significant implications for robotics, manufacturing, and autonomous systems.