© IE Online Media Services Pvt Ltd
Tags:
Journalism of Courage
Runway has taken a major step in the intensifying race to build world models, announcing its first system, GWM-1, alongside a significant upgrade to its flagship Gen 4.5 video model. The dual release signals the American tech company’s push to position itself at the front of an industry quickly shifting from experimental research to deployable, production-grade tools.
The new GWM-1 model marks Runway’s formal entry into a field that several AI labs and Big Tech companies have been accelerating toward. These systems aim to learn an internal simulation of how the real world works, enabling AI to reason, plan, and act without encountering every scenario in the training data.
Runway says its approach focuses on frame-by-frame prediction, giving the model a sense of physics, movement, and environmental continuity.
Runway’s progress on its video systems has been closely watched since the debut of Gen 4.5 earlier this month, a model that topped both Google and OpenAI on the Video Arena leaderboard. The company now describes GWM-1 as more “general” than Google’s Genie-3 and other world-model contenders, framing it as a foundation for training agents across robotics, life sciences, and additional applied domains.
As part of the launch, the company unveiled three specialised versions of the system: GWM-Worlds, GWM-Robotics, and GWM-Avatars. Although they currently function as separate models, Runway says it ultimately plans to merge them into a unified world model.
GWM-Worlds allows users to create interactive spaces from a prompt or image reference, generating environments with an understanding of geometry, lighting, and physics. These simulations currently run at 24 fps and 720p, making them suitable for early gaming use cases or for training digital agents to navigate physical-world conditions.
GWM-Robotics targets synthetic training data for robots. By simulating shifting weather, obstacles, and unpredictable conditions, Runway says it can help developers explore where robots may misinterpret instructions or deviate from safety norms. The company added that it is already in discussions with several robotics firms and enterprises and plans to open access through an SDK.
Meanwhile, GWM-Avatars focuses on lifelike digital humans, an area crowded with startups such as D-ID, Synthesia, and Soul Machines, as well as experiments inside Google. Runway intends these avatars to support applications in communication, training, and behavioural simulation.
Also Read: ‘One small step for LLMs’: Why training the first AI model in space is a breakthrough
Alongside the world-model rollout, Runway also pushed a major upgrade to its Gen 4.5 video system. The refreshed version now supports native audio, character-consistent long-form generation, and multi-shot storytelling, allowing users to create up to one-minute sequences with dialogue, ambient sound, and complex camera transitions. Existing audio tracks can be edited or replaced, and multi-shot videos of any length can be stitched and refined directly within the model.
These additions bring Gen 4.5 closer to the all-in-one video pipelines offered by rivals such as Kling, especially around audio-native production. The updated model is now available to all Runway paid-plan users.
Runway’s full GWM-1 suite appears designed to serve as both a research platform and a tool for real-world industries. With interest growing across robotics, gaming, simulation, and virtual production, the company’s latest releases suggest that world models and the video systems that enable them are quickly moving from theory into practice.
