Encord released a massive open multimodal dataset and introduced EBind, a training methodology that enables multimodal AI models to be trained on a single GPU within hours instead of days or weeks. The dataset spans text, images, audio, video, and 3D point clouds. Encord claims its method achieved performance rivaling models 17× larger by focusing on data quality and parameter efficiency. The release targets democratizing multimodal AI development, especially for smaller teams and startups.