AIFLUX 3 Action launches as an open-weight model for robot control
BFL says teams can fine-tune the 7-billion-parameter model for specific robots and tasks, with native Hugging Face LeRobot integration and deployment on Nvidia Jetson.
TLDR
BFL announced FLUX 3 Action, a model that predicts video and actions together, alongside weights, code, fine-tuning recipes and reproducible examples. The company claims first place on the RoboLab benchmark, outperforming the previous best open model by 6.1 percentage points while using 56% fewer parameters and running up to 3.95 times faster. Teams can adapt it using their own demonstrations, BFL says, with LeRobot integration developed together with Nvidia.
