Researchers at MIT and the University of British Columbia have built a single artificial intelligence model that simulates many different substances, from sliding sand and flowing water to cloth and folding proteins. Called WorldParticle, it was announced by MIT News on 9 November 2026; the research is published in ACM Transactions on Graphics and will be presented at SIGGRAPH Asia.

Until now, each material has needed its own simulator, built from scratch around the equations governing that substance. WorldParticle replaces that patchwork with one transformer model, the architecture behind ChatGPT, which learns the underlying dynamics instead of being programmed with an equation for each case.

Everything becomes particles

The model represents every material as thousands or millions of tiny three-dimensional dots, or particles, and predicts where each moves one step at a time. Before each step it separates out external forces, such as a wave slamming into a seawall, then folds them back in to correct its predictions, letting it adapt to materials and forces it never saw in training.

A merging technique compresses the particles while keeping fine detail. The approach covers six kinds of physics: cloth, elastic solids, Newtonian and non-Newtonian fluids, granular materials and molecular dynamics.

Faster on proteins, truer on sand

WorldParticle captured a collapsing sand pile where baseline methods left particles floating in the air, and modelled changing protein structures about 100 times faster than numerical simulators, while beating them on accuracy.

Wider uses could include soft robot manipulation, RNA evolution and astrophysical events such as merging galaxies. Peter Yichen Chen, a former MIT postdoc now at UBC and a co-corresponding author, put the ambition plainly: “We want to have a GPT moment for simulation.” The paper notes one trade-off: the unified model gives up some per-phenomenon accuracy against specialist solvers.