A vision-focused model built for promptable visual segmentation, capable of identifying and masking objects in both images and videos. It sits in the middle of the SAM2 family — larger than the small variants but not as heavyweight as the large version — making it a practical balance between capability and resource use.