Reka previews Rho 1 for text video and robotic actions
The research preview combines visual generation and understanding within shared context.
Reka has announced a research preview of Rho 1, a model with 19 billion parameters that combines text, images, video and robotic actions. The October 5 announcement describes shared context across understanding and generation.
A shared model with limits
In Reka’s demonstration, one conversation creates a scene, marks an object, animates it and changes the weather. The company also shows simulated robot action outputs. These are company demonstrations.
Reka acknowledges drifting scene structure, unreliable grounding across video and brittle edits. Native video is limited to 672 by 384 pixels.
Access is through contact with Reka. The announcement provides no public weights download or open API. Builders still need access and independent task testing to assess deployment readiness.
ByteForward also covered Reflection’s limited Beam preview, whose announcement separately described planned future weight access.
Illustrative photograph “black soft box” by Sven Brandsma on Unsplash, published April 17, 2019 under the Unsplash License. This photograph shows studio lighting equipment and is not an image created by Rho 1. The site converted the supplied JPEG to WebP without local edits.
