World model

A world model is an AI model that learns a representation of how an environment evolves over time — enabling planning, simulation, and prediction inside the model itself rather than in real environments.

World models entered consumer attention with video models like Sora and Veo (which implicitly learn how physical scenes evolve) and game-world models like Genie. In 2026 the research direction is: can a model trained on enough video and interaction data learn a general representation of the physical world it can plan and simulate inside? Applications: robotics planning, game world generation, autonomous driving simulation, scientific simulation. Limitations remain — current world models hallucinate physics outside their training distribution — but the trajectory is clear, and several 2026 papers show measurable progress on physical reasoning via world model intermediates.

When to use world model

• Robotics planning inside simulation.
• Game world generation.

Common mistakes

• Treating current world models as physics simulators — they hallucinate edge cases.

FAQ

Related terms

• Multimodal model — A multimodal model accepts more than one input type — text plus images, audio, or video — and reasons across them in a single forward pass.
• Diffusion model — A diffusion model is a generative neural network that creates images, video, or audio by iteratively denoising random noise toward a learned target distribution.
• Reasoning model — A reasoning model is an LLM trained to produce extensive internal chain-of-thought before its final answer, trading latency for higher accuracy on hard problems.

Last updated: 2026-06-01. Raw markdown: https://promtable.com/glossary/world-model.md.