Reka Releases Rho-1 19B Omni-Reasoning Model
Reka introduces Rho-1, a 19B parameter omni-reasoning model handling text, images, video, and robot actions within a single network.
Stock photo for illustration only, not from the actual event
- Reka has officially released Rho-1, a 19B model trained entirely from scratch.
- It reads and generates text, images, video, and robot actions over a shared KV cache.
- A distilled variant is capable of generating a 5.3-second video clip in about one second.
- The release is currently a research preview with public model weights not yet available.
Reka has made a significant stride in artificial intelligence by unveiling Rho-1, a brand-new 19-billion parameter omni-reasoning model. This advanced system has been trained completely from scratch to deliver unprecedented multimodal capabilities within a unified architecture.
The core innovation of Rho-1 lies in its single-network design, which allows it to seamlessly read and generate text, images, and videos while directly outputting robotic control actions. All of these diverse modalities operate efficiently over a shared KV cache.
Stock photo for illustration only, not from the actual event
In terms of generation speed, a distilled variant of the model proves exceptionally fast, capable of producing a 5.3-second video clip in approximately one second, highlighting its potential for responsive deployment.
Integrating text, vision, video generation, and robotic control into a single 19B network marks a crucial shift toward unified multimodal foundation models. This approach reduces architectural friction and paves the way for more cohesive autonomous robotic systems in real-world environments.
Interested developers and researchers should note that Rho-1 is currently offered solely as a research preview. Reka has not yet released the public model weights for wider deployment.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment