FLUX 3 Action leads RoboLab-120 with a 7B robot-control model
Black Forest Labs has released FLUX 3 Action, a 7B open-weights World Action Model for robot control. It processes camera images, robot-state data and a text instruction, then jointly predicts future video frames and the next sequence of robot actions. On RoboLab-120, it recorded a 42.92% task-success rate, ranking ahead of the listed alternatives.
RoboLab-120 contains 120 tabletop tasks in Isaac Sim, evaluated through 10 trials on a DROID-style Franka setup. FLUX 3 Action exceeded Cosmos 3 Nano’s 36.8% by 6.1 percentage points while using 56% fewer parameters. In a separate blind test by Positronic Robotics, it completed 28 of 30 attempts on 10 DROID tasks; Cosmos 3 Nano completed 27.
BFL provides base, guidance-distilled and step-distilled DROID recipes. The fastest version uses one sampling step, although its success rate is lower. The policy produces 32 actions at 15 Hz per call. BF16 deployment needs about 32 GB of GPU memory, while FP8 quantization and text-encoder offload allow use on 24 GB cards. The FLUX Kommunity License permits non-commercial use. Applications must enforce velocity, force and workspace limits because the model does not include them.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.