Updates from Reka AI
Building multimodal models and infrastructure for the physical AI era, enabling real-time intelligence for robotics, media, and defense.
- Rho-1: Collapsing the multimodal stack (Product launch · October 5, 2026)
Reka introduced Rho-1, a 19B parameter omni reasoning model designed as a single network capable of seeing, reasoning, acting, and rendering within a unified architecture. - Reka Inverse Dynamics Model for Interactive World Models (Project update · October 2, 2026)
The research lab unveiled the Reka Inverse Dynamics Model (RIDM), a compact system that extracts low-level motor commands from video to help train interactive world models. - Evolution of LLMs: Omni-World Models (Company update · September 17, 2026)
Reka detailed its development of omni world models, which move beyond standard multimodal systems by unifying diverse inputs and outputs into a single, shared representation. - Beyond Recognition: How Our Models Reason About Video (Project update · July 30, 2026)
Reka showcased how its models perform spatial grounding, temporal reasoning, and identity inference to understand complex events in video footage at the edge.