Reka AI's omni-model Rho-1 handles text, images, video, and robot control in a single model
decoder ·
Reka AI has released a research preview of Rho-1, a 19-billion-parameter omni-model capable of processing and generating text, images, video, and robot control actions within a single neural network. Unlike traditional systems that rely on specialized models or external tool calls, Rho-1 handles all modalities as tokens in one shared context window. The model can generate continuous video in real time and adapt to new instructions without needing to restart. To address the scarcity of robot training data, Reka AI developed an inverse dynamics model that extracts control signals from standard internet videos. The same weights used for predicting camera images are also utilized to drive robotic movements. Training required 320 H100 GPUs over a period of approximately three months. This release follows Reka AI's previous launch of Reka Core in April 2024, which competed with major multimodal language models. The development aligns with the broader industry trend toward creating comprehensive world models.
decoder