World model
A world model is an AI model that learns how an environment works well enough to predict what happens next, often from video or sensor data, so an agent can plan or train inside the simulation.
By Divya Sharma · Checked Nov 15, 2026 · How we review
The term has a precise origin and a looser modern use, so it helps to know which one a speaker means.
The research meaning
In reinforcement learning, a world model is a learned simulator. A 2018 paper by Ha and Schmidhuber trained a network to build a compressed picture of a game environment and how it changes over time. A small controller then learned to act using that compressed view, and it could even be trained inside the model's own imagined version of the game before being moved back to the real one. The point is that the agent learns from predictions instead of costly real trials.
The current product meaning
Labs now use "world model" for large generative models that produce explorable video or 3D scenes and react to actions, aimed at robotics, self-driving and games. Examples reviewed on this site include NVIDIA Cosmos 3 and DeepMind's Genie 4. These are usually trained on large amounts of video and are a kind of foundation model.
A contested use
People also argue about whether large language models contain an internal "world model" of the facts they describe. That is a claim about what a model has learned inside, not a type of product, and there is no agreed test for it.
How it differs
A text-to-video model makes a clip from a prompt and stops. A world model is meant to be stepped forward: given the current state and an action, it predicts the next state, so something can act inside it. Its limits are the same as any simulator: errors build up over long rollouts, and a policy that works in the model may not work in the real world.
Related terms
- Multimodal model
- A multimodal model is an AI model that can take in, and sometimes produce, more than one kind of data, such as text, images, audio or video, within a single model rather than through separate systems.
- Foundation model
- A foundation model is a large model trained on broad data at scale that can be adapted to many different downstream tasks, through prompting or fine-tuning, instead of being built for one task.
- Text-to-video
- Text-to-video is the task of generating a short video clip from a written prompt. Models must produce frames that each look right and also move consistently from one frame to the next.
- Large language model
- A large language model (LLM) is a neural network with billions of parameters, trained on large amounts of text to predict the next token, which lets it write, answer questions, summarise and follow instructions.
- AI agent
- An AI agent is a system in which a language model decides its own next steps: it calls tools, reads the results and repeats until the task is done, instead of following a fixed script.