Compared to Chinese and English, research on Mongolian sentiment analysis is relatively limited and remains in its developmental stage. To address the issues of information loss, weak emotional feature extraction, and insufficient modality interaction in Mongolian Multimodal Sentiment Analysis (MSA), this paper proposes a Mongolian MSA model based on a Gated Transformer and Adaptive Hyper-modality (GTAH). First, a Gated Transformer Encoder (GTE) is introduced, which filters information to prevent the loss of relevant data and enhances the ability to extract emotional features. Next, an Adaptive Hyper-modality Learning module based on Bidirectional Cross-modal Attention (BiC-AHL) is designed, which strengthens the interaction between textual and audiovisual modalities, thereby improving modal collaboration. Experimental results show that the proposed model outperforms several advanced models, effectively improving the accuracy of Mongolian MSA.