How AI describes UNI

UNI is the largest pretrained vision encoder for histopathology, built on ViT-L/16 with DINOv2, trained on around 100 million images across 100,000 whole slide images, and developed by Mahmood Lab AI for Pathology at Harvard/BWH. It serves as a feature extractor and backbone for downstream pathology AI tasks, enabling evaluation and fine-tuning for varied histology analyses, with training data drawn from private histology collections rather than public datasets. Access is gated by Hugging Face terms; the model is released under CC-BY-NC-ND-4.0 for non-commercial academic use with attribution, and commercial use requires prior approval and institutional verification.