Reka AI's 19B omni-model Rho-1 handles text, images, video and robot control in one net
The Decoder · rss · 2026-10-06
Reka AI unveiled Rho-1, a 19-billion-parameter omni-model that processes and generates text, images, video, and robot control actions in a single neural network.
- Unified architecture: no task routing to specialized systems — all modalities run as tokens in one shared context window.
- Cheap training: built on 320 H100 GPUs over roughly three months, a fraction of the compute top models use.
- Robot control is a first-class output, unusually folding embodied action generation into the omni-model framework.
More from Embodied
- Cornell's UMA robot model learns new tasks from 25 human videos with frozen weights — qinzytech · 2026-10-06
- Dyna Robotics Launches Taku, a Humanoid Robot That Does Laundry, Folding and Chores Autonomously — JasonMa2020 · 2026-10-06
- Robotics luminary Sergey Levine to headline South Park Commons fireside chat — adityaag · 2026-10-06
- Capstan Drive Prototype Shows Zero Slippage Ahead of Gimbal Motor Test — IanPritchard · 2026-10-06
- Anybotics launches Shift, an ops engine already running across 200+ robot deployments — lukas_m_ziegler · 2026-10-06
- Figure trained its F.02 robots to jump into molten steel to retire the whole fleet — lukas_m_ziegler · 2026-10-06