Prov-GigaPath (Microsoft, Providence)

Prov-GigaPath is an open pathology foundation model from Microsoft and Providence, published in Nature in 2024. It pairs a DINOv2 tile encoder with a slide-level LongNet encoder whose dilated attention can model an entire gigapixel slide, and it was trained on 1.3 billion image tiles from 171,189 slides drawn from more than 30,000 Providence patients. The weights are open, and the model performs strongly on mutation prediction and cancer subtyping benchmarks. Its limitations are that the training data come from a single US health system, which may limit generalisation to other scanners and populations, and that whole-slide attention is computationally heavy. It sits alongside UNI, CONCH and TITAN as one of the reference models of computational pathology. For a newcomer, Prov-GigaPath is a model that reads an entire slide at once rather than piece by piece.