Model families
Compare representation, action conditioning, spatial output, and availability without collapsing unlike evidence into one score.
15 models
DreamerV3
A latent-dynamics reinforcement-learning system that learns behavior from imagined trajectories.
Meta AI · Jun 11, 2025V-JEPA 2
A predictive video representation model adapted for action-conditioned robot planning.
Google DeepMind · Aug 5, 2025Genie 3
A research preview for interactively generated video worlds controlled in real time.
NVIDIA · Jun 2, 2026Cosmos 3
An open family of world models that jointly processes and generates language, images, video, audio, and actions.
World Labs · Sep 1, 2026Atlas
A spatial world model reported to generate persistent multimodal 3D worlds and video.
Runway · Dec 11, 2025GWM-1
Runway's general world model research spanning simulated environments and robotics.
Runway · Sep 3, 2026GWM Worlds 2
Runway's research preview of a real-time interactive world model generating 720p video and 48 kHz audio.
Odyssey · Sep 15, 2026Odyssey-3
Odyssey's foundation world model presented as one backbone driving robots, vehicles, drones and game agents.
Waymo · Feb 6, 2026Waymo World Model
A driving-focused generative simulator for diverse and counterfactual road scenes.
Wayve · Dec 2, 2025GAIA-3
Wayve's multi-view generative world model for driving simulation and evaluation.
Meta FAIR · Feb 15, 2024V-JEPA
Meta's first Video Joint Embedding Predictive Architecture for self-supervised latent video learning.
NVIDIA · Jan 6, 2025Cosmos 1.0
NVIDIA's first world foundation model platform for physical AI development.
Wayve · Sep 29, 2023GAIA-1
Wayve's original generative world model for autonomous-driving video synthesis.
Decart & Etched · Oct 31, 2024Oasis
A real-time interactive video world model that responds to keyboard and mouse input.
Decart · Date unknownOasis 3
An interactive world model reported to generate real-time, controllable, multi-view environments for physical-AI training.