Runway is testing Praxis-1, a robot-control model built on its video pretraining, with selected partners including Noble Machines, Standard Bots and Ultra. The announcement is dated September 2026 without a specific day and says public model weights will follow in the coming months.
The company presents demonstrations across different robot bodies and environments, alongside a comparison of video-pretraining approaches. These are company-reported results, not an independent validation of general-purpose reliability. The announcement’s broad ambitions should be distinguished from the hardware, tasks and conditions actually evaluated. A promised open-weight release is also different from weights that developers can already download.
NAOAT analysis: Evaluate the task, the recovery path and the integration cost
For an integrator, a useful trial starts with a repeatable task and a clear definition of completion. Picking up an object, placing it within a specified tolerance and recovering after a failed grasp are separate outcomes. Reporting each one makes it possible to distinguish an attractive demonstration from a dependable workflow.
The trial should include the variations expected at the intended site. Lighting, object placement and surrounding clutter can be changed systematically while keeping the task definition fixed. Recording unsuccessful attempts alongside successful ones helps reveal where the model needs additional data, a revised fixture or a simpler operating procedure.
Integration deserves its own budget. A policy that produces plausible actions still needs compatible sensors, control interfaces, timing and monitoring. Teams should account for the work required to connect those components and establish a safe fallback when the task cannot be completed. Those are proposed evaluation requirements, not capabilities demonstrated by the announcement.
When weights become available, licensing terms, supported hardware and reproducible evaluation instructions will help developers compare options. Until then, partner trials can be useful evidence, provided their conditions remain visible. The business question is how much reliable work a complete installation delivers for the effort required to deploy and maintain it.