MeshXL is a generative model family designed for high-fidelity 3D mesh generation. The paper introduces Neural Coordinate Field (NeurCF), an explicit coordinate representation with implicit neural embeddings, which facilitates the auto-regressive generation of 3D meshes. NeurCF encodes 3D coordinates into a sequence of tokens, making it suitable for large-scale mesh modeling. MeshXL leverages modern large language model approaches to train these models on a collection of over 2.5 million 3D meshes from various datasets. Extensive experiments demonstrate that MeshXL can generate high-quality 3D meshes, outperforming existing methods in terms of diversity, quality, and scalability. The model is also effective for conditional 3D asset generation, given additional inputs such as images or text. MeshXL's performance improves with larger model sizes and more training data, making it a strong foundation for various downstream applications.MeshXL is a generative model family designed for high-fidelity 3D mesh generation. The paper introduces Neural Coordinate Field (NeurCF), an explicit coordinate representation with implicit neural embeddings, which facilitates the auto-regressive generation of 3D meshes. NeurCF encodes 3D coordinates into a sequence of tokens, making it suitable for large-scale mesh modeling. MeshXL leverages modern large language model approaches to train these models on a collection of over 2.5 million 3D meshes from various datasets. Extensive experiments demonstrate that MeshXL can generate high-quality 3D meshes, outperforming existing methods in terms of diversity, quality, and scalability. The model is also effective for conditional 3D asset generation, given additional inputs such as images or text. MeshXL's performance improves with larger model sizes and more training data, making it a strong foundation for various downstream applications.