Reka launches Rho-1 with 19 billion parameters for video and robot actions

First reported by RuntimeWire at · Updated · 2 sources

According to MarkTechPost, the network was trained from scratch and processes and generates text, images, video and robot actions using a shared KV cache. The publication says a distilled variant produces a 5.3-second clip in about one second. RuntimeWire reports training used 320 H100 GPUs over three months, while the research preview retains short-horizon and resolution limits.

Covered by 2 publishers within 5 hours of the first report.

Reporting2

RuntimeWire Reka releases a 19B model for video generation and robot control · Ryan Merket
MarkTechPost Reka Releases Rho-1: A 19B Omni-Reasoning Model That Understands, Generates Video and Outputs Robot Actions in One · Michal Sutter