AI world model progress
Read the field as a path from creating and exploring worlds to controlling, simulating, and building with them.
From clip to placeMain path
The goal is orientation, not a paper tracker: first what a person can do, then which company and model lane supports that action.
The clearest change is that some systems now have product surfaces visitors can open, create with, and judge visually.
The useful question is not which model wins. It is what each company actually lets a visitor control, inspect, or compare.
Beyond visible products, the category is moving toward reusable assets, simulation, physical AI, and inhabited worlds.
Action map
Create and Build are the product-facing lanes. Explore, Control, and Simulate explain the research and adjacent progress without making it sound like every model is already a public tool.
Create persistent 3D worlds from prompts, images, video, or spatial inputs for scenes that can be explored and reused.
Understand how keyboard actions can steer an AI-generated world frame by frame instead of replaying a fixed clip.
See how promptable interactive environments differ from ordinary generated video or hand-authored game levels.
Explore Alibaba's product-facing path for directing an open world while it continues unfolding.
Generate 360-degree environment worlds for VR backdrops, games, mood boards, and web 3D scenes.
Inspect Tencent's open 3D world generation and reconstruction path for persistent scene assets and engine-oriented workflows.
Study what happens when many AI agents inhabit a shared world, specialize, organize, and form social patterns.
Already usable
World Labs / Marble and Skybox AI are the clearest consumer-facing surfaces in the current site story. They stay visible, but they are not turned into a station for orders, payments, or automatic generation.
Create editable 3D worlds in Marble.
Generate 360 environments with Skybox AI.
