Will it run?
Products

Runway launches agent with asset tagging, reorganizes around general world model

By Marco Vane Clawpit staff
Runway launches agent with asset tagging, reorganizes around general world model

Runway has unveiled Runway Agent with a tagging feature that lets users point the agent directly at assets in their workspace — video, images, audio — and request edits, alternatives, or fixes. At the same time, the company is restructuring its entire offering around what it calls "real-world intelligence," a general world model named GWM-1 designed to understand, simulate, and operate in the physical world.

The agent and workflow

The centerpiece is Agent Tagging: users can tag the agent wherever their assets live and ask it to make changes or generate alternative versions. Runway did not release comparative benchmarks for the agent, nor did it detail which models power it beyond a general statement that it rests on the same "real-world intelligence" infrastructure.

Three platforms, one foundation

The company now divides its product into three arms. Runway Creative is the full creative suite, an all-in-one cloud platform for creating and editing video, images, and audio, aimed at individuals and teams of any size. Runway Dev is an API platform for developers that exposes Runway's models and those from other labs, with built-in editing tools for production-scale work. Runway Robotics targets robotics: a toolkit for running policy inference through photorealistic simulation, synthetic data generation, and evaluation, all driven by GWM-1.

GWM-1 and Gen-4.5: what's behind the names

GWM-1 is presented as an advanced general world model built for interaction with the real world, what the company calls a major step toward universal simulation. Gen-4.5 is described as the world's best video model, with top-tier motion quality, prompt adherence, and visual fidelity; Runway provided no public benchmarks or numerical comparisons against Sora, Veo 2, or other competing models. The General World Models project is framed as a long-term research effort to build AI systems that understand the visual world and its dynamics.

What's missing from the announcement

No official availability date for Agent Tagging beyond "try today," no pricing details for the three platforms, and no technical specification for GWM-1 — size, training data, architecture, or known limitations. The company also did not clarify whether the models will be available as open weights or only as a managed service. Until independent data appears, the announcement remains at the level of intent statements and a repackaging of an existing product line.