To address the demand for real-time monitoring of snow leopards, this paper proposes a lightweight snow leopard detection algorithm, You Only Look Oncev8-Global Slimming (YOLOv8-GS), while maintaining high detection accuracy. First, a lightweight feature extraction module, Grouped Channel Partition Convolution Attention (GCPCA), is designed. By employing channel slicing and depthwise separable convolutions, GCPCA reduces both parameter count and computational cost, and incorporates an Efficient Channel Attention (ECA) mechanism to enhance key channel feature representation. In addition, a parameter-free Simple Attention Mechanism (SimAM) is integrated into the backbone network, which amplifies important pixel information by computing the deviation of each pixel from the mean across channels without increasing computational overhead, thereby improving detection performance. A dedicated snow leopard image dataset is constructed for model training and evaluation. Experimental results demonstrate that YOLOv8-GS achieves an mAP@0.5 of 94.1% on the self-built dataset, with a parameter count of 2.00M, 5.8 GFLOPs, and a model size of 4.43 MB. Compared with the baseline YOLOv8 model, mAP@0.5 is improved by 2.3%, while the parameter count, computational cost, and model size are reduced by 33.3%, 29.27%, and 25.8%, respectively.