Reka AI releases 19B omnimodal model Rho-1, single model handles text, images, video and robotic control

Reka AI releases research preview of 1.9 billion parameter omnimodal model Rho-1, processing and generating text, images, video and robotic control actions within a single neural network. The model puts all modalities as tokens into the same context window, without tool calls or external models, can generate continuous video in real time and respond instantly to new instructions, and the same set of weights both predicts camera images and drives robot motion.