Physics-IQ audit says a third of prompts were ambiguous and 30% of videos inflated scores

DynamicWebPaige · x · 2026-07-25

A new Physics-IQ audit found that claims about “physically accurate” video models are only as good as the benchmark behind them. More than one-third of prompts were ambiguous, and about 30% of videos contained artifacts that inflated scores.

After cleaning up the benchmark, the model rankings changed materially, with a reported τ = 0.46. The takeaway: vibes ≠ physics when evaluating video generation systems.

Related event: Physics-IQ Audit Reveals Flaws in Video Model Benchmarks(2 posts)→

Original post →

More from Multimodal

Multimodal channel →