Black Forest Labs has introduced an open-weight 7B world-action model that operates 3.95x faster than the strongest competing open vision-language-action models. The model, called FLUX 3 Action, takes first place on the RoboLab benchmark, outperforming the previous best open model by 6.1 percentage points while using 56% fewer parameters. It predicts video and actions simultaneously and plans more than twice as far ahead as previous open alternatives.

The model is integrated into Hugging Face's LeRobot for edge deployment on Nvidia Jetson hardware. Black Forest Labs is releasing the model weights, code, benchmarks, and training recipes to enable fine-tuning for specific robot policies and task-specific simulated environments, including gaming, vehicle control, and computer use.