Black forest labs unveils flux 3 action, a 7b-parameter model leading robolab-120

Black Forest Labs (BFL) has launched FLUX 3 Action, a 7B open-weights World Action Model (WAM) designed for robot control. The model processes camera frames, robot state, and text instructions to predict future video frames and actions, achieving a 42.92% task success rate on the RoboLab-120 leaderboard. It outperforms NVIDIA’s Cosmos 3 Nano, which ranks second at 36.8%, and surpasses π0.5, which scores 28.0%. FLUX 3 Action uses a smaller backbone and distillation to close the speed gap, offering real-time performance with a 1.34x to 2.28x advantage over π0.5 in FP8 on workstation and datacenter GPUs. The model is pretrained on a mix of image, video, and audio data, with video comprising over 95% of training tokens. It supports edge deployment on NVIDIA Jetson and integrates with Hugging Face LeRobot. BFL also tested hybrid control with GPT 6 Astra, achieving 90% episode success at a lower cost. The model outputs joint targets without built-in velocity or force limits, requiring application-specific constraints.