The paper introduces a new benchmark to evaluate the quality of generated videos from the Sora model, focusing on their adherence to real-world physics principles. The authors transform the generated videos into 3D models and use the fidelity of geometric constraints satisfied by these models as a proxy for the videos' compliance with physical laws. They compare Sora's performance with other models like Pika and Gen2, demonstrating that Sora outperforms them in terms of geometry consistency. The evaluation is based on metrics such as the number of correct matching points, retention ratio, and geometric reprojection error. The results show that Sora's videos exhibit higher geometric consistency and sustained stability compared to the other models, highlighting its ability to produce videos with realistic and physically accurate content. The paper also discusses future work directions, including the need for more comprehensive assessment tools that consider additional physics-based metrics.The paper introduces a new benchmark to evaluate the quality of generated videos from the Sora model, focusing on their adherence to real-world physics principles. The authors transform the generated videos into 3D models and use the fidelity of geometric constraints satisfied by these models as a proxy for the videos' compliance with physical laws. They compare Sora's performance with other models like Pika and Gen2, demonstrating that Sora outperforms them in terms of geometry consistency. The evaluation is based on metrics such as the number of correct matching points, retention ratio, and geometric reprojection error. The results show that Sora's videos exhibit higher geometric consistency and sustained stability compared to the other models, highlighting its ability to produce videos with realistic and physically accurate content. The paper also discusses future work directions, including the need for more comprehensive assessment tools that consider additional physics-based metrics.