Black Forest Labs reports FLUX 3 ranks first on Google DeepMind's Physics-IQ benchmark

Black Forest Labs announced that FLUX 3 ranked first on Google DeepMind's Physics-IQ benchmark, an evaluation suite designed to test physical world understanding in video models. The benchmark tests whether a model can predict what physically happens next given the opening frames of a filmed experiment, covering fluid dynamics, optics, solid mechanics, magnetism, and thermodynamics.

According to Black Forest Labs, FLUX 3 substantially outperforms competing frontier models in most evaluation cases, including Seedance 2.5, MiniMax H3, Gemini Omni 1.1 Flash, Veo 3.1, Sora 2, and Cosmos3. The FLUX 3 base model placed ahead of all other evaluated foundation models, and pairing the model with a physics verification layer yielded higher scores.

The company stated that the underlying physics understanding in the FLUX 3 backbone powers action prediction in its robotics model FLUX 3 Action and realistic motion in FLUX 3 Video.