SenseTime releases NEO architecture, redefining the performance boundaries of multimodal models.

SenseTime, in collaboration with Nanyang Technological University, released NEO, the world's first native multimodal architecture, breaking through the traditional patchwork paradigm of "visual encoder + projector + language model". The architecture achieves deep integration of vision and language through three major innovations: native tile embedding, 3D rotational position encoding, and native multi-head attention.