Reka has introduced Rho-1, a groundbreaking omni-reasoning model that boasts 19 billion parameters and is trained from the ground up. This innovative model integrates the ability to read and generate text, images, video, and control robot actions, all utilizing a shared key-value cache.
The model’s capabilities are exemplified by its distilled variant, which can produce a 5.3-second video clip in approximately one second. Currently, Rho-1 is available as a research preview, but it does not yet have public weights for broader use.
This launch marks a significant milestone in AI development, showcasing Reka’s commitment to advancing the field with versatile models that can perform multiple tasks across different media formats. As AI continues to evolve, models like Rho-1 could pave the way for more integrated and intelligent systems in various applications.
Compiled automatically by the Tech AI Newsdesk from public AI-news sources and summarised in our own words.
