August 18, 2025

DINOv3 is a next-generation vision foundation model trained purely with self-supervised learning. It introduces innovations that allow robust dense feature learning at scale with models reaching 7B parameters and achieves