Welcome, Praxis 1!

Praxis 1 is Runway’s first robot foundation model, a software system positioned at the intersection of video learning and robotics. Runway, the New York AI company known for video generation tools, describes the project as a “world action model.” Its development strategy centers on video as a training resource, including ordinary clips from the web, rather than relying primarily on collections of robot demonstrations. That choice reflects the company’s view that video offers a more scalable route to gathering training material for robotics. The model has a Humanoid Guide Brain Score of 4 out of 10, placing it in the developing range rather than among mature or leading systems. Its significance lies in the direction of the research: connecting a company’s existing investment in video models with the demands of robotics software. Praxis 1 should therefore be understood as a developing foundation model initiative, not as a physical robot or a finished hardware product. The emphasis is on a shared software foundation for future robotic applications.

For robotics teams, the central technical question is how to evaluate a foundation model separately from the machine that uses it. A useful assessment distinguishes the software, the training approach, the integration work, and the hardware environment. Video centered training also raises practical questions about dataset composition, evaluation coverage, and how results should be compared with approaches built around robot demonstrations. These are evaluation priorities, not established performance claims for Praxis 1. Prospective users should define their intended application, select measurable success criteria, and examine evidence under the conditions relevant to that application. A developing score makes careful benchmarking particularly important before committing to deployment.

Runway announced Praxis 1 in September 2026 and plans a public release with open weights in the following months. Named early partners include Noble Machines, Standard Bots, and Ultra, placing the project within an emerging network of robotics collaborators rather than solely within Runway’s existing video software business. The listing marks availability as backorder, while the planned public release remains a future milestone. An open weights release would give researchers and engineering teams a more direct basis for examining the model and assessing its suitability for their own work. More broadly, the project represents Runway’s expansion into robotics and its effort to make video training central to that strategy. Partnership activity and release plans establish the project’s direction, while its developing score keeps expectations grounded.

 

Similar Posts