AI research startup Black Forest Labs has launched FLUX 3 Action, an open-weights world-action model designed specifically for robotics applications. Built upon the multimodal FLUX 3 architecture, the model processes multi-camera video feeds from a robot’s workspace to predict immediate physical actions and simulate subsequent environmental changes.
According to Black Forest Labs, the 7-billion-parameter model set a benchmark record on the RoboLab-120 evaluation leaderboard for task success rate. FLUX 3 Action is less than half the size of previous leading open robotics models while executing up to 3.95 times faster, addressing the latency and footprint constraints that often prevent large reasoning models from running on edge hardware.
The model weights have been uploaded and made publicly available on Hugging Face. Beyond physical hardware deployments, the company noted potential applications in digital environments, including autonomous navigation in gaming and high-speed computer-use agents.
Why it matters
Provides hardware and robotics developers with an open, high-performance world model optimized for edge deployment.
Reduces compute latency and memory overhead for real-time robotic action planning and simulation.
Expands multimodal world-model architectures into computer control and gaming environments.
Source: the-decoder.com



