Aether AI releases 16B CausalWM, a world model that reasons about causality before generating future frames

jiqizhixin · x · 2026-09-29

Conventional world models take the current frame plus an action command and directly generate future video, but motion, contact, and spatial change stay compressed inside the model — you can't tell whether it understood the physics or merely memorized plausible appearances.

Aether AI has released the first version of CausalWM, a 16B-parameter causal world model introducing a paradigm it calls Causal Chain-of-Thought: instead of rushing to generate the future, the model first reasons out how objects move and how 3D geometric relations change, then generates future frames from those intermediate results — explicitly writing out the causal process, aiming to shift world models from predicting what the future looks like to predicting how it happens step by step.

Original post →

More from Models

Models channel →