Physics-IQ
Physical prediction in generated video across mechanics, fluids, optics, thermodynamics and magnetism. Original and Verified results use separate protocols.


Rather
The benchmark catalog
Start with the capability you care about. Explore what each benchmark measures, which models it covers, and where the results come from.
Physical plausibility, everyday common sense, and predicting what happens next.
3 suites · 150 published results
Categories group related tests. Each suite keeps its own metrics and protocols; category counts include only suites with published results.
Physical prediction in generated video across mechanics, fluids, optics, thermodynamics and magnetism. Original and Verified results use separate protocols.


Multi-turn interactive world-model evaluation across quality, setting, interaction, consistency and physical behavior.


Instruction following, physics adherence and common sense in video generation models, with native scores rather than a rescaled universal ranking.

