Lorentz-Equivariant Geometric Algebra Transformers for High-Energy Physics This paper introduces the Lorentz Geometric Algebra Transformer (L-GATr), a new architecture for high-energy physics that is equivariant under Lorentz transformations. L-GATr represents data in geometric algebra over four-dimensional space-time and is a Transformer, making it versatile and scalable. It is demonstrated on regression and classification tasks from particle physics and is used to construct the first Lorentz-equivariant generative model based on a continuous normalizing flow. L-GATr is on par with or outperforms strong domain-specific baselines across various tasks, including quantum field theory amplitude prediction, top tagging, and generative modeling of reconstructed particles. The architecture is designed to handle the rich structure of high-energy physics data, including the Lorentz symmetry of special relativity. L-GATr supports variable-length inputs and is efficient in training, making it suitable for large systems. The paper also discusses the computational cost and scalability of L-GATr, showing that it performs well in terms of data efficiency and scalability. The results show that L-GATr is a versatile architecture that can be applied to a wide range of tasks in high-energy physics.Lorentz-Equivariant Geometric Algebra Transformers for High-Energy Physics This paper introduces the Lorentz Geometric Algebra Transformer (L-GATr), a new architecture for high-energy physics that is equivariant under Lorentz transformations. L-GATr represents data in geometric algebra over four-dimensional space-time and is a Transformer, making it versatile and scalable. It is demonstrated on regression and classification tasks from particle physics and is used to construct the first Lorentz-equivariant generative model based on a continuous normalizing flow. L-GATr is on par with or outperforms strong domain-specific baselines across various tasks, including quantum field theory amplitude prediction, top tagging, and generative modeling of reconstructed particles. The architecture is designed to handle the rich structure of high-energy physics data, including the Lorentz symmetry of special relativity. L-GATr supports variable-length inputs and is efficient in training, making it suitable for large systems. The paper also discusses the computational cost and scalability of L-GATr, showing that it performs well in terms of data efficiency and scalability. The results show that L-GATr is a versatile architecture that can be applied to a wide range of tasks in high-energy physics.