Black Forest Labs enters robotics with open model FLUX 3 Action

Black Forest Labs is entering robotics with FLUX 3 Action, an open-world-action model that uses camera feeds to predict what action a robot should take next.

The move marks a significant expansion for a company best known for its work on image-generation models. With FLUX 3 Action, Black Forest Labs is applying its modeling expertise to physical control — a domain where AI systems must translate visual input into decisions about movement and manipulation in real time.

The model does this with just seven billion parameters. That relatively compact size matters. According to the announcement, FLUX 3 Action sets a record on the RoboLab-120 benchmark while running up to 3.95 times faster than the previous top model.

What the model does

FLUX 3 Action is an open-world-action model. It takes camera feeds as input and predicts the next action a robot should take. This framing positions the model as a decision-making layer for robotics: rather than handling perception or language alone, it outputs actionable predictions grounded in what the robot's cameras see.

The emphasis on "open-world" signals that the model is designed to operate beyond tightly controlled or pre-mapped environments, where robots historically struggle to generalize. Predicting the next action from raw visual input is one of the central problems in modern robot learning, and it is the specific problem FLUX 3 Action targets.

Benchmark performance and efficiency

Two numbers stand out in the announcement. First, FLUX 3 Action sets a record on RoboLab-120, the benchmark on which the previous top model also held its lead. Second, it runs up to 3.95 times faster than that previous top model.

The combination is the story. A seven-billion-parameter model that is both more accurate and dramatically faster than the prior leader challenges the assumption that robotics foundation models need to be enormous to compete. Speed carries particular weight in robotics: a model that predicts actions faster can support more responsive control loops, and responsiveness is a practical requirement for robots operating in dynamic, real-world settings.

Open weights

Black Forest Labs describes FLUX 3 Action as an open model. The company has built its reputation on openly released models, and it is now extending that approach to robotics. Open availability matters here for a concrete reason: it allows robotics researchers and developers to inspect, fine-tune, and build on the model directly rather than relying on closed APIs, which has historically been the default for state-of-the-art robot-control systems from large labs.

If FLUX 3 Action's benchmark results hold up under independent evaluation, an open model at the top of RoboLab-120 could shift how robotics teams source their control models — and put pressure on labs that keep their strongest robot models proprietary.

Why it matters

Robotics is widely viewed as the next major application domain for large-scale AI models, but the field has been constrained by models that are either too slow for real-time control or too closed for broad adoption. A seven-billion-parameter open model that claims both a benchmark record and a 3.95x speed advantage over the previous leader addresses both constraints at once.

For Black Forest Labs, the release also represents diversification beyond image generation into a market with substantial commercial stakes. The announcement gives no pricing or deployment details, but the benchmark claims alone position the company as an immediate contender in robot-action modeling.

The next question is independent verification: whether FLUX 3 Action's record on RoboLab-120 and its speed advantage translate into performance on real robot hardware outside benchmark conditions.