Runway publishes its first global model and updates its video model with native audio
Runway Launches GWM-1 World Model, Expands AI Simulation and Video Capabilities
AI image and video generation company Runway has officially entered the world-model race with the launch of its first world model, GWM-1. The company says the model creates realistic simulations through frame-by-frame prediction, allowing it to understand physics, motion, and how the real world behaves over time.
What Is a World Model?
A world model is an AI system that learns how the world works internally so it can reason, plan, and act without being trained on every real-life scenario. Runway says GWM-1 is more “general” than Google’s Genie-3 and other competing models.
Earlier this month, Runway released its Gen 4.5 video model, which topped Google and OpenAI on the Video Arena leaderboard. Now, with GWM-1, Runway aims to build simulations that can train AI agents for robotics, science, and interactive environments.
Runway’s Approach
Runway’s CTO Anastasis Germanidis said the company first needed an advanced video model to build a strong world model:
“The right path is teaching models to predict pixels directly. At sufficient scale and with the right data, you can build a model that understands how the world works.”
Three Versions of GWM-1
Runway introduced three specialized versions of the world model:
1. GWM-Worlds
An interactive tool that lets users create a virtual environment using a prompt or reference image. The simulation runs at 24 fps and 720p, with realistic geometry, physics, and lighting. Beyond gaming, Worlds can be used to teach agents how to navigate and behave in real-world conditions.
2. GWM-Robotics
Designed for robotics training using synthetic data enriched with variables like weather changes, obstacles, and environment shifts. Runway says this can help identify how and when robots might break rules or fail in unexpected scenarios.
3. GWM-Avatars
A system for creating realistic digital humans. Several companies—including D-ID, Synthesia, Soul Machines, and Google—have built similar technologies for training, communication, and interactive uses. Runway plans to eventually merge Worlds, Robotics, and Avatars into one unified model.
Gen 4.5 Receives a Major Upgrade
Runway also updated its Gen 4.5 model with:
Native audio generation
Long-form, multi-shot video creation
Up to one-minute videos with consistent characters
Background audio, dialogue, and multi-angle shots
Editing tools for both audio and video
This update brings Runway closer to competitors like Kling, which also released a powerful all-in-one video suite this month. It shows that video-generation models are rapidly evolving into production-ready filmmaking tools.
Gen 4.5’s updated version is now available to all paid plan users.
Availability and Industry Interest
Runway says GWM-Robotics will be offered through an SDK and that the company is already in discussions with robotics firms and enterprise customers about using both Robotics and Avatars.
With GWM-1, Runway positions itself at the forefront of next-generation AI simulation—where video, virtual worlds, and intelligent agents converge into a single, powerful ecosystem.