Black Forest Labs is releasing FLUX 3 Action. This is an open AI model designed specifically for robotics.
The new software builds on the multimodal FLUX 3 model. That foundational model was trained mainly on video along with image and audio data.
FLUX 3 Action functions as a world-action model. It ingests multi-camera video feeds originating from a robot workspace.
From those feeds, the system predicts the next action an agent should take. It also forecasts how the surrounding environment will change as a result.
The company states that the model achieves a record success rate on the RoboLab-120 leaderboard. This performance comes with just seven billion parameters.
The model is less than half the size of the previous best open model. It runs up to 3.95 times faster than that predecessor.
Large reasoning models excel at planning. However, they frequently prove too slow and too bulky for physical robots.
This efficiency makes the technology critical for on-device deployment. Black Forest Labs also identifies potential applications inside digital environments.
Video games can act as testing grounds for navigation capabilities. Fast-reacting agents that operate computers could follow those developments down the line.
The model weights are available on Hugging Face.
