The rapid growth of short video platforms has accelerated the dissemination of fake news in the form of videos. Compared to traditional fake news, short video fake news integrates multi-modal information such as text, audio, and video. However, existing fake news detection methods face significant challenges in processing long-sequence data like videos and uncovering the inherent relationships between diverse modalities. To address the above problems, we propose a short video fake news detection based on MambaFormer method called SVFDMF. Specifically, SVFDMF alternates between mamba layers, multi-head attention mechanisms, and adaptive normalization layers to effectively extract long-sequence features from modalities such as video and audio while preserving temporal structural information. In addition, we design a Transformer-based cross-modal fusion mechanism to capture more complex inter-modal correlations. Extensive experimental results demonstrate that our proposed method outperforms existing methods. The code is available at https://github.com/SVFDMF-main/SVFDMF .