MedMamba is a novel vision Mamba model designed for generalized medical image classification. It integrates convolutional layers for local feature extraction with state space models (SSMs) for long-range dependency modeling, enabling efficient and accurate classification of medical images from various modalities. The model employs a hybrid basic block called SS-Conv-SSM, combining grouped convolution and channel-shuffle operations to reduce computational complexity and parameter size. MedMamba was evaluated on 16 datasets containing 411,007 images across ten imaging modalities, demonstrating competitive performance compared to state-of-the-art methods. The model achieves high accuracy, precision, sensitivity, specificity, and AUC, with MedMamba-B outperforming other models in most metrics. MedMamba's efficient architecture, with low FLOPs and parameter size, makes it suitable for practical applications in medical imaging. The model's robustness to various image perturbations and its ability to provide interpretable results through Grad-CAM and t-SNE visualization further highlight its effectiveness. MedMamba establishes a new baseline for medical image classification and offers valuable insights for developing more powerful SSM-based AI systems in healthcare. The source code and pre-trained weights are available at https://github.com/YubiaoYue/MedMamba.MedMamba is a novel vision Mamba model designed for generalized medical image classification. It integrates convolutional layers for local feature extraction with state space models (SSMs) for long-range dependency modeling, enabling efficient and accurate classification of medical images from various modalities. The model employs a hybrid basic block called SS-Conv-SSM, combining grouped convolution and channel-shuffle operations to reduce computational complexity and parameter size. MedMamba was evaluated on 16 datasets containing 411,007 images across ten imaging modalities, demonstrating competitive performance compared to state-of-the-art methods. The model achieves high accuracy, precision, sensitivity, specificity, and AUC, with MedMamba-B outperforming other models in most metrics. MedMamba's efficient architecture, with low FLOPs and parameter size, makes it suitable for practical applications in medical imaging. The model's robustness to various image perturbations and its ability to provide interpretable results through Grad-CAM and t-SNE visualization further highlight its effectiveness. MedMamba establishes a new baseline for medical image classification and offers valuable insights for developing more powerful SSM-based AI systems in healthcare. The source code and pre-trained weights are available at https://github.com/YubiaoYue/MedMamba.