ModelsSelected

Runway Releases Praxis-1, a 27B MoE Video-Action Diffusion Model

Why we picked it

A video-generation company moving from text-to-video toward unified world models suggests the next competitive front is interactive simulation, not prettier clips. Builders in gaming, robotics and simulation should read the technical report to assess whether a 27B MoE design offers usable control fidelity. Compared with DeepMind's Genie line, Praxis-1's open technical detail makes it a practical reference for architecture choices.

Runway announced Praxis-1, a 27B-parameter mixture-of-experts diffusion transformer that jointly predicts video and actions. The company published a technical report on its research page detailing the model's architecture, training stages and results. Praxis-1 combines video generation and action prediction in a single model, rather than treating world modelling and video synthesis as separate tasks.

Image · Runway ML
Body

The model is built on a diffusion transformer backbone with a mixture-of-experts design at 27B parameters. By predicting video and actions together, Praxis-1 is aimed at interactive and controllable generation, where the model must infer what happens next in a scene and what actions drive that outcome. Runway's technical report covers the training pipeline across multiple stages, though the announcement itself provides limited headline benchmark numbers.

World models that couple action prediction with video generation are becoming a competitive area, with work from Google DeepMind (Genie) and Chinese labs like Alibaba and Tencent in the same space. For Runway, a video-generation company best known for Gen-3, Praxis-1 signals a move toward interactive simulation and potentially robotics-adjacent applications. Developers and enterprises building simulation, gaming or embodied-AI products can review the full technical report on Runway's site.

Original · Runway ML

Praxis-1 is a 27B-parameter mixture-of-experts diffusion transformer that predicts video and actions together. We detail the model's architecture, training stages and results in our technical report. runway.com/research/praxi…📊 2 ⚡

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Runway, Praxis-1, world models, video generation.

Source attribution: curated from Runway ML (@runwayml).