Reka Rho-1

Reka Rho-1 is a research model that combines work on text, images, video and actions in one system. Reka announced the 19-billion-parameter proof of concept on 5 October 2026.

The research combines several kinds of output

The project explores whether one model can reason across information types that are often handled by separate systems. Its demonstrated video output uses a native resolution of 672 × 384. The published examples and comparisons describe Reka’s own research evaluation.

An integrated model could reduce the number of separate components needed in an interactive application. Its usefulness still depends on whether those components work reliably together on the intended task, including the consistency of generated frames and the interpretation of actions.

Rho-1 is a proof of concept

Reka presents Rho-1 as a research preview. The announcement presents research demonstrations and invites collaboration enquiries. It establishes a proof of concept with a contact route for collaborators.

The companion Reka inverse dynamics model has a separate downloadable release. That model estimates camera movement from video. Its research announcement explains the narrower task and should be read separately from Rho-1’s broader multimodal work.