Companies and AI labs

Companies Building World Models

Meet the companies and research teams behind world-model products, public demos, open-source systems, and physical-AI platforms. Each profile separates what the organization ships from what remains research.

04

Supported here

Models connected to the World Models Watch generators.

Public product · Usable product

Create editable 3D worlds in Marble.

Create, edit, explore, share, and export persistent 3D worlds.

Models

  1. Marble

    Creates persistent 3D worlds from images, text, video, spatial layouts.

  2. World API

    World Labs' developer platform for generating and manipulating 3D worlds.

Public product · Usable product

Generate 360 environments with Skybox AI.

Create skyboxes, HDRI-style environments, GLB exports, and scene backdrops.

Model

  1. Skybox AI

    Generates 360-degree environments: skyboxes, HDRI lighting, VR scenes, game-engine backplates.

Public product · Usable product

Turn images into explorable 3D worlds with Echo.

Open the Echo product, provide an image, and evaluate a generated 3D world through the company's hosted surface.

Model

  1. Echo-2

    Turns text, images, or panoramas into persistent 3D worlds that can be explored and shared in a browser.

Local open source · Open stack

Build a navigable indoor world with SpatialGen.

Inspect open checkpoints and examples, or use Manycore's product ecosystem to understand image-to-room and spatial-design workflows.

Model

  1. SpatialGen

    Generates spatially consistent indoor 3D scenes from an image or a text-conditioned layout workflow.

01

Use now

Public products you can open and use today.

Public product · Product signal

Runway shows video tools moving toward characters and worlds.

World-model research direction plus a deployed character-agent product surface.

Model

  1. GWM-1

    Runway's general world model spanning generated worlds, avatars, robotics-oriented directions.

07

Try a demo

Public or limited previews for hands-on evaluation.

Public demo · Experimental preview

Oasis is a playable realtime AI-world demo.

A keyboard-controlled demo world showing interaction beyond fixed clips.

Model

  1. Oasis

    Realtime open-world model: interactive video world from keyboard input, frame-by-frame.

Public demo · Experimental preview

Step into a live AI video world with Odyssey-2.

Open a streamed AI world, enter text while it is running, and watch the scene respond in real time.

Model

  1. Odyssey-2

    A live generated video world that responds to text while the stream is running.

Public demo · Open stack

Run a playable Waypoint world locally through Biome.

Try Waypoint in a browser, or download Biome and move through generated environments on supported local hardware.

Model

  1. Waypoint 1.5

    A real-time generated world that can run in a browser or locally on supported consumer hardware.

Public demo · Experimental preview

Play with world-and-action generation through WHAM-RT.

Open the official WHAM page, follow the Copilot Labs entry, and interact with a research-scale generated gameplay scene.

Model

  1. WHAM-RT

    A real-time gameplay model that generates visuals in response to controller input.

Limited access · Product signal

HappyOyster is an early look at interactive AI worlds.

Consumer-style interface for directing and exploring AI worlds; limited access.

Model

  1. HappyOyster

    Alibaba's open-ended world model for real-time immersive creation, exploration, directing.

Limited access · Experimental preview

Genie shows where prompt-based playable worlds are heading.

Interactive demos plus a gated prototype with Street View-grounded scenes.

Model

  1. Genie 3

    Frontier world model generating and navigating interactive environments from prompts.

Limited access · Product signal

Watch MoWorld turn captured scenes into interactive 3D spaces.

Read the official launch materials and watch examples of interactive 3D reconstruction, editing, and new-view generation.

Model

  1. MoWorld 3D

    A world-model platform for dynamic scene reconstruction, editing, novel views, and physical-AI simulation.

06

Open source & research

Local releases and research-first systems without a public product.

Local open source · Open stack

Tencent makes the open 3D asset track easier to follow.

Public repos, model cards, reports for 3D worlds, embodied AI.

Models

  1. HY-World 2.0

    Reconstructs, generates, simulates persistent 3D worlds from text, images, video.

  2. HY-Embodied-0.5

    Tencent's embodied foundation-model family: spatial reasoning, planning, robot perception, 0.5-X.

Local open source · Open stack

Robbyant connects world models with maps and robot control.

Open repos, papers, model releases spanning worlds, maps, robot control.

Models

  1. LingBot-World

    Robbyant's open-source world simulator for high-fidelity, long-horizon, real-time interactive environments.

  2. LingBot-Map

    Feed-forward 3D model for streaming video reconstruction; public KITTI/Oxford-Spires benchmarks.

  3. LingBot-VA

    Robbyant's causal video-action world model for generalist robot control.

  4. LingBot-VLA

    Robbyant's pragmatic vision-language-action model for generalist robotic manipulation across platforms.

Local open source · Adjacent progress

Cosmos explains the physical-AI infrastructure lane.

Platform story: synthetic data, world foundation models, robot-training workflows.

Model

  1. Cosmos

    Physical-AI world foundation model platform; Cosmos 3 arrived March 2026.

Local open source · Open stack

Explore the open Matrix-Game interactive world model stack.

Watch official examples, inspect the repository, and run the released model locally with keyboard and mouse controls.

Model

  1. Matrix-Game 3.0

    A streaming keyboard-and-mouse world model designed for real-time interaction and long-horizon memory.

No consumer access · Adjacent progress

Project Sid shows how AI inhabitants behave inside a world.

Minecraft-based simulation where agents specialize, organize, form social patterns.

Model

  1. Project Sid

    Minecraft-based many-agent simulation: specialization, law, culture, religion in AI civilizations.

No consumer access · Open stack

Understand physical prediction and planning with V-JEPA 2.

Read the official explanation, inspect open artifacts, and view demonstrations of physical prediction and robot planning.

Model

  1. V-JEPA 2

    A video-trained latent world model for understanding, predicting, and planning actions in the physical world.