This paper proposes a Hash-grid Assisted Context (HAC) framework for compressing 3D Gaussian Splatting (3DGS) models. The HAC framework leverages the inherent spatial consistencies of unorganized anchors through a structured hash grid to achieve highly compact 3DGS representations. The key contributions include: (1) being the first to model contexts for 3DGS compression using a structured hash grid to exploit the inherent consistencies among unorganized 3D Gaussians; (2) proposing to use interpolated hash features to neural-predict the value distribution of anchor attributes for efficient entropy encoding; and (3) introducing an adaptive quantization module and learnable masks to enhance compression efficiency. The HAC framework achieves a compression ratio of 11× over the base model Scaffold-GS and 75× over the vanilla 3DGS model, while maintaining comparable or improved fidelity. The framework is evaluated on five datasets, demonstrating significant size reduction and improved performance. The HAC framework is implemented based on the Scaffold-GS model, incorporating a structured hash grid for context modeling and an adaptive quantization module for efficient entropy coding. The framework also includes an adaptive offset masking strategy to eliminate invalid Gaussians and anchors. The results show that the HAC framework achieves state-of-the-art compression performance for 3DGS models.This paper proposes a Hash-grid Assisted Context (HAC) framework for compressing 3D Gaussian Splatting (3DGS) models. The HAC framework leverages the inherent spatial consistencies of unorganized anchors through a structured hash grid to achieve highly compact 3DGS representations. The key contributions include: (1) being the first to model contexts for 3DGS compression using a structured hash grid to exploit the inherent consistencies among unorganized 3D Gaussians; (2) proposing to use interpolated hash features to neural-predict the value distribution of anchor attributes for efficient entropy encoding; and (3) introducing an adaptive quantization module and learnable masks to enhance compression efficiency. The HAC framework achieves a compression ratio of 11× over the base model Scaffold-GS and 75× over the vanilla 3DGS model, while maintaining comparable or improved fidelity. The framework is evaluated on five datasets, demonstrating significant size reduction and improved performance. The HAC framework is implemented based on the Scaffold-GS model, incorporating a structured hash grid for context modeling and an adaptive quantization module for efficient entropy coding. The framework also includes an adaptive offset masking strategy to eliminate invalid Gaussians and anchors. The results show that the HAC framework achieves state-of-the-art compression performance for 3DGS models.