The paper introduces a novel framework called Hash-grid Assisted Context (HAC) for compressing 3D Gaussian Splatting (3DGS) models, which are used for novel view synthesis. 3DGS models are known for their rapid rendering speed and high fidelity but require significant storage space due to the large number of Gaussians and their attributes. The HAC framework leverages the structured hash grid to exploit the inherent spatial consistencies among unorganized Gaussians, which are often sparse and unorganized. By using a binary hash grid, the framework establishes continuous spatial consistencies and uses a context model to predict the value distributions of anchor attributes, facilitating entropy coding. The approach also includes an Adaptive Quantization Module (AQM) to dynamically adjust quantization steps for different attributes and an adaptive masking strategy to eliminate invalid Gaussians and anchors. The HAC framework achieves a significant reduction in model size, over 75× compared to vanilla 3DGS, while maintaining or improving fidelity. Extensive experiments on multiple datasets demonstrate the effectiveness of the HAC framework, showing a notable improvement over existing 3DGS compression methods.The paper introduces a novel framework called Hash-grid Assisted Context (HAC) for compressing 3D Gaussian Splatting (3DGS) models, which are used for novel view synthesis. 3DGS models are known for their rapid rendering speed and high fidelity but require significant storage space due to the large number of Gaussians and their attributes. The HAC framework leverages the structured hash grid to exploit the inherent spatial consistencies among unorganized Gaussians, which are often sparse and unorganized. By using a binary hash grid, the framework establishes continuous spatial consistencies and uses a context model to predict the value distributions of anchor attributes, facilitating entropy coding. The approach also includes an Adaptive Quantization Module (AQM) to dynamically adjust quantization steps for different attributes and an adaptive masking strategy to eliminate invalid Gaussians and anchors. The HAC framework achieves a significant reduction in model size, over 75× compared to vanilla 3DGS, while maintaining or improving fidelity. Extensive experiments on multiple datasets demonstrate the effectiveness of the HAC framework, showing a notable improvement over existing 3DGS compression methods.