Rho-1
AI model · Robotics / world model · Text → Text by Reka AI
A 19B omni reasoning model that sees, reasons, renders, and acts.
Rho-1 is a 19B parameter omni reasoning model for physical AI that unifies seeing, reasoning, rendering, and acting in one architecture. It is designed for researchers and developers building autonomous systems that require a deep understanding of real-world physics and motion.
Model specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
- Launched
- 2026-10-05
- Languages
- English
Category: Generative AI
- Domains
- Robotics, autonomous systems, physical AI
- Parameters
- 19B
- Access
- Research preview via Reka Cloud
- Released
- 2026-10-05
- Model Type
- Omni Reasoning Model
- Input Modalities
- Language, Image, Video, Actions
- 19B parameter omni reasoning model
- Unified network that sees, reasons, renders, and acts
- Designed for physical-world intelligence and robotics
- Processes discrete and continuous signals in one architecture
Overview
Rho-1 is a 19-billion parameter omni reasoning model developed by Reka as a research preview for the era of physical AI. It is built on a single unified architecture that natively processes and generates multiple modalities, including text, images, video, and actions, without relying on an orchestration of separate agents. Designed for robotics and autonomous systems, Rho-1 aims to solve the gap in physical-world intelligence by understanding motion, geometry, and the laws of physics. The model integrates both discrete signals like language and continuous signals like video pixels or robotic arm configurations into a shared representation. This unified approach reduces the latency and cost associated with model-swapping in traditional agentic chains. Rho-1 serves as a foundational world model that can simulate, reason about, and interact with the physical environment in real time.
Physical Intelligence
- Unified Omni Architecture: The model is designed as a single unified network that natively sees, reasons, renders, and acts within one architecture.
- 19B Parameter Scale: Rho-1 is a 19-billion parameter model built to handle the complexity of physical-world intelligence.
- Integrated Generation and Understanding: The model unifies generation and understanding, avoiding the latency and cost of swapping between separate agents for different modalities.
- Continuous and Discrete Signal Processing: It is engineered to handle both discrete signals like language and continuous signals like video pixels or arm joint configurations.
Reasoning and Simulation
- World Modeling: The model is built to understand the rules of motion, geometry, and physical laws required to operate in the real world.
- Physical Simulation: Rho-1 serves as a research preview for systems that can simulate and reason about the physical world.
- Shared Representations: It uses shared representations where image, video, and language tokens are projected into an interconnected space for simultaneous reasoning.
Use cases
- Researchers use Rho-1 to develop autonomous systems that can reason about physical laws and motion in real time.
- Developers leverage the unified architecture to build multimodal applications that generate actions and video without agent hand-off latency.
- Robotics teams apply the model's shared representations to process continuous sensor data and discrete commands in a single network.
Specifications
- Modalities
- Text, image, video, audio, actions → Text, image, video, audio, actions
- Domains
- Robotics, autonomous systems, physical AI
- Parameters
- 19B
- Access
- Research preview via Reka Cloud
- Released
- 2026-10-05
- Model Type
- Omni Reasoning Model
- Input Modalities
- Language, Image, Video, Actions
- Output Modalities
- Language, Image, Video, Actions
- Primary Focus
- Physical AI and World Modeling
FAQ
What is Rho-1?
Rho-1 is a 19-billion parameter omni reasoning model that unifies seeing, reasoning, rendering, and acting in a single network.
How does Rho-1 differ from standard Vision-Language Models (VLMs)?
Unlike typical VLMs that only output language, Rho-1 is designed to output multiple modalities including images, videos, and actions.
What is the advantage of Rho-1 over agentic AI chains?
Rho-1 is an omni model that processes multimodal reasoning within a single shared representation, whereas agentic AI chains different models together, incurring hand-off latency and higher costs.
Added by AntSpace from public sources (reka.ai).