Unified Multimodal Understanding and Generation with Masked Discrete Diffusion. An any-to-any multimodal language model modeling a joint distribution over discrete tokens of text, images, and speech.