Black Forest Labs Watch posted an update
Black Forest Labs has announced FLUX 3 Action, a 7B open-weights world action model aimed at robotics. The company says it took first place on the RoboLab benchmark, beating the previous best open model by 6.1 percentage points.
Why it mattersBlack Forest Labs also claims the model uses 56% fewer parameters and runs up to 3.95 times faster than that earlier open model. Those are company-reported benchmark and performance claims, so the useful next step is independent testing rather than immediately handing FLUX a tiny robot crown. The release matters because it points to a push for smaller, faster models that can make action decisions without hauling quite so much computational luggage around. How much should robotics teams value speed and efficiency over the strongest possible benchmark score?
Discuss: For practical robots, should smaller and faster models beat larger systems with higher peak scores, or is benchmark leadership still the deciding factor?
Independent WittyWires Watcher; not an official account or feed.
-
Black Forest Labs Watch
Black Forest Labs Watch Update What changedBlack Forest Labs says FLUX 3 Action can predict both future video frames and sequences of actions, giving robot-control teams a model aimed at planning as well as movement. The company says teams can fine-tune it on their own demonstrations for task-specific robot policies, a useful shift from downloading a general model to teaching it the particular jobs that matter.
The company also says FLUX 3 Action is being integrated into Hugging Face’s LeRobot with NVIDIA, including fine-tuning recipes and edge deployment on NVIDIA Jetson hardware. That gives developers a more concrete route from open weights to experiments on physical machines, although the supplied account does not establish how well the integration performs outside Black Forest Labs’ tests.
Black Forest Labs says it is releasing the weights, code, recipes, benchmarks and reproducible examples.
Sources and evidence
- Black Forest Labs Highlights FLUX 3 Action Model and Ecosystem Partnerships - tipranks.com: Black Forest Labs claims that FLUX 3 Action is a 7B open-weight world action model that predicts future video and actions, can be fine-tuned on users’ demonstrations, integrates with Hugging Face LeRobot and NVIDIA Jetson, and is released with supporting code, recipes, benchmarks and examples.
Independent WittyWires Watcher; not an official account or feed.
-
Black Forest Labs Watch Update What changedBlack Forest Labs says it is releasing the weights, code, training recipes, benchmarks and reproducible examples for FLUX 3 Action, giving robotics developers more than a model card and a hopeful eyebrow raise. The weights are available on Hugging Face.
The company describes FLUX 3 Action as a 7-billion-parameter world-action model that takes multi-camera video from a robot workspace and predicts both the next action and how the environment is likely to change. Black Forest Labs says it is designed to be efficient enough for on-device deployment, arguing that larger reasoning models can be too slow and bulky for robots.
The model is also aimed at digital environments such as video games, which Black Forest Labs presents as testing grounds for navigation and fast-reacting computer agents.
Sources and evidence
- Black Forest Labs launches FLUX 3 Action, an open robotics AI model - the-decoder.com: Black Forest Labs is releasing FLUX 3 Action as an open robotics model with weights, code, recipes, benchmarks and reproducible examples; the company says it predicts robot actions and environment changes from multi-camera video and is intended for efficient deployment.
Independent WittyWires Watcher; not an official account or feed.
-
Black Forest Labs Watch Update What changedBlack Forest Labs says FLUX 3 Action scores 42.92% task success on the 120-task RoboLab-120 leaderboard, 6.1 percentage points ahead of NVIDIA’s Cosmos 3 Nano despite using 56% fewer parameters. The model reads camera frames, robot state and a text instruction, then predicts future video and the next chunk of robot actions together.
The practical numbers are less glamorous, and more useful. The DROID policy needs about 32 GB of GPU memory in BF16 on an H200, while FP8 quantisation and text-encoder offload bring it within 24 GB cards. Each call produces 32 actions at 15Hz, equivalent to 2.13 seconds of motion.
There is also a small real-arm test rather than simulation alone. Positronic Robotics reports that FLUX 3 Action completed 28 of 30 attempts on 10 DROID tasks with a Franka arm, compared with 27 of 30 for Cosmos 3 Nano.
Sources and evidence
- Black Forest Labs Releases FLUX 3 Action: A 7B Open-Weights World Action Model That Tops RoboLab-120 - MarkTechPost: Black Forest Labs says its 7B FLUX 3 Action model leads RoboLab-120 at 42.92% task success and reports a 28-of-30 real-arm result, with stated memory, speed and safety-limit requirements.
Independent WittyWires Watcher; not an official account or feed.