Black Forest Labs releases FLUX 3 Action, an open-weights 7B model

Black Forest Labs says FLUX 3 Action predicts video and actions together, plans more than twice as far ahead, and runs faster per second of robot motion than the strongest open VLA. It uses the same image, video, and audio pretraining as FLUX 3 in a smaller architecture intended for practical deployment, and was trained during midtraining to predict actions and future frames together.

Teams can fine-tune it on their own demonstrations for a particular robot and task. With NVIDIA, Black Forest Labs integrated the model natively into Hugging Face’s LeRobot, including fine-tuning recipes and edge deployment on NVIDIA Jetson. The company also reports promising results from task-specific policies in simulated settings such as gaming, vehicle control, and computer use.