What are world models?
They try to represent how an environment changes and what consequences particular actions have.
What is it?
World models try to represent how an environment works: what happens if you move an object, turn a car or drop a ball. They let a system try out actions in simulation before carrying them out. They matter for robotics, self-driving and video models.
An everyday example
A robot that “imagines” whether a glass will slip before picking it up is using a simple world model.
Why does it matter?
Researchers see them as a route to systems that not only talk but also plan and act safely in the physical world.
A common misconception
That a simulated future is guaranteed. A world model is an approximation and can be wrong, especially in unusual situations.
In more detail
How are they different from language models?
The focus is on the state of the world, how it changes and what could happen next — not just on sequences of words.
Where are they being researched?
In robotics, simulations, games and planning.
Are they an exact copy of the world?
No. Every model is a simplification and can fail in unexpected situations.
Why do they matter for agents?
Because to act safely, it helps to estimate the consequences before the system does anything.
Remember: A simulated future is a model, not a guaranteed event.
Reliable sources
Last reviewed 1 November 2026. An explainer, not professional, legal or medical advice. For rules and figures that change, check the official version.