SenseTime Open Sources SenseNova U1, Achieving a Multimodal Native Unified Architecture

SenseTime has released and open-sourced the SenseNova U1 series of models, based on its self-developed NEO-unify architecture, achieving deep unification of multimodal understanding, reasoning, and generation, marking a transition from an integrated approach to a native unified one. The architecture discards the modular design, eliminating visual encoders and variational autoencoders, thereby improving model efficiency and performance.