FLUX 3 tops Google DeepMind’s test of physical-world video reasoning

FLUX 3 ranked first on Google DeepMind’s Physics-IQ benchmark, putting an image-generation lab at the top of a test for physical-world video reasoning. The benchmark shows models the start of a filmed experiment and asks them to predict what happens next across fluid dynamics, optics, solid mechanics, magnetism, and thermodynamics.

Black Forest Labs said FLUX 3 outperformed Seedance 2.5, MiniMax H3, Gemini Omni 1.1 Flash, Veo 3.1, Sora 2, and Cosmos3. The base model led foundation models, while a physics-verification layer scored higher.