This paper introduces SuperCluster, a novel and efficient method for large-scale 3D panoptic segmentation. The approach redefines the task as a scalable graph clustering problem, enabling efficient processing of large point clouds without the need for resource-intensive instance-matching steps during training. SuperCluster is trained using local auxiliary tasks, making it highly efficient and adaptable to the superpoint paradigm, which further enhances its scalability and performance. The model processes scenes with millions of points and thousands of objects in a single inference, achieving state-of-the-art results on two indoor scanning datasets: 50.1 PQ (+7.8) for S3DIS Area 5 and 58.7 PQ (+25.2) for ScanNetV2. It also sets the first state-of-the-art for two large-scale mobile mapping benchmarks: KITTI-360 and DALES. With only 209k parameters, SuperCluster is over 30 times smaller than the best-competing method and trains up to 15 times faster. The model's code and pretrained models are available at https://github.com/drprojects/superpoint_transformer. The method is particularly resource-efficient, fast, and scalable, with high precision, as demonstrated in Figure 1. The primary contributions include large-scale panoptic segmentation, fast and scalable segmentation, and the extension to superpoints for enhanced scalability. The paper also presents experiments on various datasets, showing that SuperCluster achieves high semantic segmentation performance and sets new state-of-the-art results on multiple benchmarks. The method is efficient, scalable, and effective for large-scale 3D panoptic segmentation.This paper introduces SuperCluster, a novel and efficient method for large-scale 3D panoptic segmentation. The approach redefines the task as a scalable graph clustering problem, enabling efficient processing of large point clouds without the need for resource-intensive instance-matching steps during training. SuperCluster is trained using local auxiliary tasks, making it highly efficient and adaptable to the superpoint paradigm, which further enhances its scalability and performance. The model processes scenes with millions of points and thousands of objects in a single inference, achieving state-of-the-art results on two indoor scanning datasets: 50.1 PQ (+7.8) for S3DIS Area 5 and 58.7 PQ (+25.2) for ScanNetV2. It also sets the first state-of-the-art for two large-scale mobile mapping benchmarks: KITTI-360 and DALES. With only 209k parameters, SuperCluster is over 30 times smaller than the best-competing method and trains up to 15 times faster. The model's code and pretrained models are available at https://github.com/drprojects/superpoint_transformer. The method is particularly resource-efficient, fast, and scalable, with high precision, as demonstrated in Figure 1. The primary contributions include large-scale panoptic segmentation, fast and scalable segmentation, and the extension to superpoints for enhanced scalability. The paper also presents experiments on various datasets, showing that SuperCluster achieves high semantic segmentation performance and sets new state-of-the-art results on multiple benchmarks. The method is efficient, scalable, and effective for large-scale 3D panoptic segmentation.