UltraLight VM-UNet is a lightweight model designed for skin lesion segmentation, significantly reducing parameters and computational complexity. Based on Vision Mamba, it achieves strong performance with only 0.049M parameters and 0.060 GFLOPs. The model introduces a Parallel Vision Mamba Layer (PVM Layer) that processes features in parallel, maintaining constant channel count while minimizing parameters. This layer uses four parallel Vision Mamba modules, each with a quarter of the original channel count, leading to a 93.7% reduction in parameters. The model also incorporates skip-connection paths with Spatial Attention Bridge (SAB) and Channel Attention Bridge (CAB) modules for multi-scale feature fusion. Experimental results on three skin lesion datasets (ISIC2017, ISIC2018, PH²) show that UltraLight VM-UNet outperforms existing lightweight models in terms of parameter efficiency while maintaining competitive performance. The model's design reduces parameter explosion by analyzing key factors affecting Mamba parameters, such as channel count, state dimension, and convolution kernel size. The PVM Layer enables efficient feature processing with minimal computational overhead, making it suitable for mobile medical devices. The model's performance is validated through extensive experiments, demonstrating its effectiveness in medical image segmentation tasks.UltraLight VM-UNet is a lightweight model designed for skin lesion segmentation, significantly reducing parameters and computational complexity. Based on Vision Mamba, it achieves strong performance with only 0.049M parameters and 0.060 GFLOPs. The model introduces a Parallel Vision Mamba Layer (PVM Layer) that processes features in parallel, maintaining constant channel count while minimizing parameters. This layer uses four parallel Vision Mamba modules, each with a quarter of the original channel count, leading to a 93.7% reduction in parameters. The model also incorporates skip-connection paths with Spatial Attention Bridge (SAB) and Channel Attention Bridge (CAB) modules for multi-scale feature fusion. Experimental results on three skin lesion datasets (ISIC2017, ISIC2018, PH²) show that UltraLight VM-UNet outperforms existing lightweight models in terms of parameter efficiency while maintaining competitive performance. The model's design reduces parameter explosion by analyzing key factors affecting Mamba parameters, such as channel count, state dimension, and convolution kernel size. The PVM Layer enables efficient feature processing with minimal computational overhead, making it suitable for mobile medical devices. The model's performance is validated through extensive experiments, demonstrating its effectiveness in medical image segmentation tasks.