ContextGS is a novel framework for compressing 3D Gaussian Splatting (3DGS) by introducing an autoregressive model at the anchor level. The method reduces spatial redundancy among anchors by dividing them into hierarchical levels, where coarser-level anchors are used to predict finer-level anchors. This approach significantly improves coding efficiency and reduces storage requirements. ContextGS achieves an average compression ratio of 15× compared to Scaffold-GS and 100× compared to standard 3DGS, while maintaining or improving rendering quality. The method introduces a low-dimensional quantized feature as a hyperprior for each anchor, enabling more efficient entropy coding. The framework also employs a multi-level anchor division strategy, where decoded anchors from coarser levels are directly used in finer levels, reducing storage overhead. The proposed method demonstrates superior performance in terms of compression efficiency and rendering quality compared to existing techniques. The results show that ContextGS achieves significant improvements in storage efficiency and rendering quality, making it a promising solution for 3DGS compression. The method is evaluated on real-world datasets and shows effective compression and rendering performance. The framework is implemented based on Scaffold-GS and includes entropy coding and masking loss to optimize the compression process. The method is validated through extensive experiments, demonstrating its effectiveness in reducing storage requirements and improving rendering quality. The results highlight the importance of context modeling in 3DGS compression and show that the proposed method achieves state-of-the-art performance in terms of compression efficiency and rendering quality.ContextGS is a novel framework for compressing 3D Gaussian Splatting (3DGS) by introducing an autoregressive model at the anchor level. The method reduces spatial redundancy among anchors by dividing them into hierarchical levels, where coarser-level anchors are used to predict finer-level anchors. This approach significantly improves coding efficiency and reduces storage requirements. ContextGS achieves an average compression ratio of 15× compared to Scaffold-GS and 100× compared to standard 3DGS, while maintaining or improving rendering quality. The method introduces a low-dimensional quantized feature as a hyperprior for each anchor, enabling more efficient entropy coding. The framework also employs a multi-level anchor division strategy, where decoded anchors from coarser levels are directly used in finer levels, reducing storage overhead. The proposed method demonstrates superior performance in terms of compression efficiency and rendering quality compared to existing techniques. The results show that ContextGS achieves significant improvements in storage efficiency and rendering quality, making it a promising solution for 3DGS compression. The method is evaluated on real-world datasets and shows effective compression and rendering performance. The framework is implemented based on Scaffold-GS and includes entropy coding and masking loss to optimize the compression process. The method is validated through extensive experiments, demonstrating its effectiveness in reducing storage requirements and improving rendering quality. The results highlight the importance of context modeling in 3DGS compression and show that the proposed method achieves state-of-the-art performance in terms of compression efficiency and rendering quality.