Short extractable summary preserved directly in static HTML.
| Model | Lab or company | Category |
|---|---|---|
| DreamerV3 | Google DeepMind | Model-Based RL |
| DreamerV2 | Model-Based RL | |
| PlaNet | Model-Based RL | |
| RSSM | Latent Dynamics | |
| NVIDIA Cosmos | NVIDIA | Foundation World Model |
| Genie 2 | Google DeepMind | Generative World Model |
| MuZero | Google DeepMind | Model-Based RL |
| Predictron | Google DeepMind | Model-Based RL |
| UniSim | Google DeepMind | Generative World Model |
| TD-MPC2 | MIT / Meta | Model-Based RL |
| World Models (Ha & Schmidhuber) | Google Brain / IDSIA | Model-Based RL |
| V-JEPA | Meta | Self-Supervised World Model |
| IRIS | Microsoft Research | Model-Based RL |
| GAIA-1 | Wayve | Foundation World Model |
| Imagination-Augmented Agents (I2A) | Google DeepMind | Model-Based RL |
| Value Prediction Network (VPN) | University of Michigan / Google Brain | Model-Based RL |
| AMI World Model | AMI Labs | Foundation World Model |
| Sora | OpenAI | Generative World Model |
| DIAMOND | Microsoft Research / University of Geneva | Model-Based RL |
| Pandora | Tsinghua University / ByteDance | Generative World Model |
| OASIS | Decart / Etched | Generative World Model |
| Copilot4D | Waabi | Foundation World Model |
| Genie | Google DeepMind | Generative World Model |
| I-JEPA | Meta | Self-Supervised World Model |
| GameNGen | Google Research | Generative World Model |
| Emu Video | Meta | Generative World Model |
| 3D-VLA | MIT / Tsinghua | Foundation World Model |
| RT-2 | Google DeepMind | Foundation World Model |
| Large World Model (LWM) | UC Berkeley | Foundation World Model |
| Stable Video Diffusion | Stability AI | Generative World Model |
| MILE | Wayve | Foundation World Model |
| STEVE-1 | UT Austin | Generative World Model |
| Gen-3 Alpha | Runway | Generative World Model |
| Genie 3 | Google DeepMind | Generative World Model |
| V-JEPA 2 | Meta | Self-Supervised World Model |
| LeWorldModel | Mila / NYU / Samsung SAIL | Self-Supervised World Model |
| PixVerse R1 | PixVerse | Generative World Model |
| Marble | World Labs | Foundation World Model |
| 1X World Model | 1X | Foundation World Model |
| PlayWorld | Princeton University | Generative World Model |
| WHAM | Microsoft Research / Ninja Theory | Generative World Model |
| WHAM-RT | Microsoft Research / Ninja Theory | Generative World Model |
| GAIA-2 | Wayve | Generative World Model |
| Waabi World | Waabi | Generative World Model |
| Odyssey-2 | Odyssey | Foundation World Model |
| HY-World 2.0 | Tencent Hunyuan | Foundation World Model |
| RELIC | Adobe Research | Generative World Model |
| Matrix-Game 2.0 | Skywork AI | Generative World Model |
| Lab | Type | Focus |
|---|---|---|
| Google DeepMind | industry | model-based-rl, generative-world-models, planning, games, robotics |
| NVIDIA Research | industry | foundation-world-models, physical-ai, autonomous-driving, robotics, video-generation |
| Meta FAIR | industry | self-supervised-learning, jepa, embodied-ai, autonomous-intelligence |
| Mila / NYU / Samsung SAIL | academic | self-supervised-learning, jepa, physical-reasoning, embodied-ai |
| AMI Labs | industry | multimodal-world-models, embodied-ai, language-conditioned-robotics, foundation-models |
| UC Berkeley | academic | model-based-rl, robotics, visual-rl, sim-to-real |
| Stanford University | academic | robotics, embodied-ai, manipulation, simulation |
| Carnegie Mellon University | academic | robotics, autonomous-systems, model-based-learning |
| Toyota Research Institute | industry | autonomous-driving, robotics, physical-ai |
| MIT CSAIL | academic | model-based-rl, planning, embodied-ai, scalable-agents |
| OpenAI | industry | video-generation, world-simulation, foundation-models, generative-ai |
| Wayve | industry | autonomous-driving, generative-world-models, end-to-end-driving, scenario-generation |
| PixVerse Research | industry | real-time-world-models, interactive-generation, multiplayer-experiences, video-generation |
| Decart | industry | real-time-world-models, interactive-generation, neural-game-engines |
| Waabi | industry | autonomous-driving, lidar-world-models, 4d-prediction, closed-loop-simulation |
| Microsoft Research | industry | autoregressive-world-models, diffusion-rl, sample-efficient-rl |
| Google Research | industry | neural-game-engines, world-simulation, video-understanding |
| Stability AI | industry | video-generation, diffusion-models, open-source-ai |
| Runway | industry | video-generation, creative-ai, controllable-generation |
| UT Austin | academic | instruction-following, open-world-agents, video-pretraining |
| World Labs | industry | 3d-generation, multimodal-world-models, persistent-worlds, spatial-reasoning |
| 1X | industry | humanoid-robotics, physical-ai, video-world-models, embodied-ai |
| Princeton University | academic | robotics, world-models, manipulation, autonomous-data-collection |
| Tsinghua University | academic | embodied-ai, generative-world-models, robotics, 3d-reasoning |
| ByteDance Research | industry | generative-world-models, video-generation, embodied-ai |
| University of Geneva | academic | model-based-rl, diffusion-models, world-models |
| IDSIA | academic | foundational-world-models, model-based-rl, generative-world-models |
| Etched | industry | real-time-world-models, interactive-generation, neural-game-engines |
| Odyssey | industry | interactive-world-models, causal-video, real-time-generation |
| Tencent Hunyuan | industry | 3d-world-models, multimodal-generation, world-reconstruction |
| Adobe Research | industry | interactive-world-models, long-horizon-memory, video-generation |
| Skywork AI | industry | interactive-world-models, real-time-generation, open-source |
| Category | Definition |
|---|---|
| Model-Based RL | Model-based reinforcement learning is an approach where agents learn a predictive model of the environment (a world model) and use it to simulate outcomes, plan ahead, and learn from imagined experience. |
| Embodied AI | Embodied AI concerns AI systems that interact with and learn from the physical world through a body, whether a robot, an autonomous vehicle, or a virtual agent with physical constraints. |
| Generative World Models | Generative world models are AI systems that learn to generate realistic simulated environments, replacing or augmenting hand-crafted simulators with learned models of world dynamics. |
| Latent Dynamics | Latent dynamics models learn compressed representations of environment dynamics in a latent (hidden) space, enabling efficient prediction and planning without operating in pixel space. |
| Autonomous Agents | Autonomous agents use world models to make independent decisions in complex, open-ended environments by predicting consequences and planning actions. |
| Video World Models | Video world models understand and generate video as a representation of world dynamics and physics, learning temporal structure, object permanence, and physical interactions from video data. |
| Foundation World Models | Foundation world models are large-scale, general-purpose models trained on massive datasets to learn broad representations of world dynamics. They aim to serve as versatile base models for diverse downstream tasks, from robotics to autonomous driving to video generation. |
| Self-Supervised World Models | Self-supervised world models learn representations of environment dynamics without explicit labels or reward signals. They leverage prediction in abstract representation space, predicting future states, video frames, or embeddings, to build internal models of how the world works. |