Black Forest Labs releases open-weights robotics model FLUX 3 Action

By Splitfeed EditorialLast updated Sep 26, 2026, 19:17 UTC

On or around September 23 to 24, 2026, Black Forest Labs, the company best known for its FLUX image generation models, released FLUX 3 Action, an open-weights AI model built for robotics. The model contains seven billion parameters and works by processing camera feeds to predict what action a robot should take next in real time. According to the available reporting, it tops the RoboLab-120 benchmark while running up to 3.95 times faster than the previous top-ranked model on that benchmark. Notably, it reaches that performance at roughly half the parameter count of its closest rivals.

The release matters for a few reasons beyond the benchmark numbers. Black Forest Labs has no prior history in robotics, so this is a meaningful expansion of its scope, and releasing the model as open weights means developers and researchers can use, inspect, and build on it without paying licensing fees or operating through a closed API. The combination of high benchmark performance, smaller model size, and open access is the specific claim that makes this announcement consequential. What remains genuinely in dispute is how well RoboLab-120 scores translate to real-world robotic performance across varied environments, and whether the speed and efficiency gains hold outside controlled benchmark conditions.

None of the three camps have published reactions to this story yet. The Pro-AI camp would typically highlight the open-weights release as a democratizing move that accelerates progress in physical AI and puts powerful robotics tools in the hands of independent researchers. The Anti-AI camp would likely raise concerns about the downstream risks of making capable robotic control models freely available without restrictions, particularly around misuse or deployment in unsafe contexts. The Middle Ground camp would probably welcome the transparency of open weights while calling for clearer evaluation of how benchmark results map onto real-world reliability and safety.

The argument will sharpen once independent researchers and robotics teams test FLUX 3 Action on hardware outside the RoboLab-120 benchmark. Reproducibility results and third-party evaluations published in the weeks following the release would be the clearest signal of whether the efficiency and performance claims hold up at scale.