DriveDreamer from Giga AI is the world first physics-world-driven autonomous driving world model. It uses diffusion models with multi-stage training to achieve complex traffic environment modeling and future prediction, and was recognized as ECCV 2024 most influential paper. DriveDreamer4D is the first generation-plus-reconstruction autonomous driving 4D world model, and the company pioneered the world-model-plus-reinforcement-learning paradigm for autonomous driving. The technology generates high-fidelity complex and long-tail driving scenarios from text descriptions and structured definitions, reducing data acquisition costs by two orders of magnitude, and has been commercially deployed with over 20 OEMs and autonomous driving companies.