GWM-1 by Runway: status and timeline

GWM-1

World model creating 3D virtual environments from text. Three variants: robotics sim (physics), avatar generation, creative use.

DeployStage 4 of 5

Three variants: world sim (physics), avatars, creative; enables virtual robot testing.

Updated 11 Dec 2025Checked 10 Oct0 updates this week

Milestones

Next · Adoption for robotics and avatar creation begins

GWM-1 released in three variantsDec 2025Complete.
Adoption for robotics and avatar creation beginsTarget 2026Current milestone.

Most important updates

Runway releases GWM-1, a world model for robotics simulation and avatar creation11 Dec 2025

Upcoming

2026Adoption for robotics and avatar creation begins (next)
2026Integration with game engines and VFX software
Q4 2026Adoption by robotics labs for algorithm testing

Current obstacles

Sim-to-real transfer accuracyRobot control trained in simulation often fails on real hardware due to unmodeled dynamics, friction and sensor noise.
Scale: generating large, complex environmentsCurrent models handle single scenes well; generating and maintaining consistency across large open worlds remains challenging.

Physics limits

Simulation fidelity is bounded by training dataThe model can only synthesize variations of physics and scenarios seen in training; novel interactions may be unrealistic.
Photorealism and physics accuracy trade offHigher visual quality sometimes means less accurate physics; optimizing both simultaneously is difficult.

How it works

3 parts

Text-to-world generation

Creates 3D scenes from natural language

GWM-1 takes a text description and generates a photorealistic 3D virtual environment, with realistic physics and lighting.

Physics-based simulation

Enables realistic robot movement testing

The world simulation variant accurately simulates forces, collisions and friction, so robots trained in simulation transfer to real hardware.

Avatar generation

Specialized variant for character creation

The avatar variant optimizes for generating realistic human and creature characters with natural motion, suitable for games and film.

Spec sheet

Spec	GWM-1
Number of specialized variants	3 (world simulation, avatar, creative)^R (reported)

^R reported by the company

Papers & demos

1 item

Dec 2025demo
GWM-1 world model release (opens runway.com)World model enabling virtual environment creation for robotics testing, neural network training and avatar generation.

Update log

1 update

Thu 11 Dec 2025

Major: Runway releases GWM-1, a world model for robotics simulation and avatar creationPressSoftware

About Runway

The team behind GWM-1

Runway

Gen-4.5 video generation, GWM-1 world model, generative media

New York-based company founded 2018 by Cristóbal Valenzuela, Anastasis Germanidis, Alejandro Matamala. Develops Gen-4.5 video generation and GWM-1 world model for robotics and avatars.

Raised $315M Series E Feb 2026 at $5.3B valuation; led by General Atlantic, Nvidia, AMD.
Gen-4.5 (Dec 2025): top-ranked video generator; excels at physics (momentum, fluids).
GWM-1 (Dec 2025): world model for robot simulation, avatar generation, virtual test environments.

runwayml.com (opens runwayml.com)Company profile →

Founded: 20188 yrs
Headquarters: United States
Status: Private
Valuation: $5.3BprivateFeb 2026
Raised: $858M6 rounds
Last round: Series E · $315MFeb 2026$5.3B post
Works in: AIImage & voice generation
Coverage: 2 programs · 3 updateslatest 10 Feb 2026checked 10 Oct
Lead investors: General AtlanticGoogleFelicisCoatueAmplify Partners
Investors: Nvidia (opens nvidia.com)General Atlantic (opens generalatlantic.com)AMD Ventures (opens amd.com)
People: Cristóbal Valenzuela (opens twitter.com)Co-founder & CEOAnastasis Germanidis (opens twitter.com)Co-founder & CTOAlejandro Matamala (opens twitter.com)Co-founder

Spec

GWM-1

Number of specialized variants

3 (world simulation, avatar, creative)^R (reported)