Black Forest Labs Launched FLUX 3 Action Robotic Model
The seven-billion-parameter open-weight model utilizes visual-audio integration for robotic control tasks.
Updated on Sept. 23, 2026 in Robotics
Live Poll
Do you trust that new AI robotic models will make local workplaces safer and more efficient?
Black Forest Labs has launched the FLUX 3 Action robotic model, a seven-billion-parameter architecture capable of integrating image, video, and audio data. The model utilizes a new technique called Self-Flow to support control sequences and video generation.
Why it matters
By leveraging a video-prediction backbone to generate robot control commands, the model enables faster fine-tuning on limited demonstration datasets. This approach addresses the high data requirements typically associated with deploying robots in complex manufacturing environments.
FLUX 3 Action achieves 80 to 101 milliseconds of end-to-end latency on consumer hardware. It requires only 30 minutes of demonstration data to fine-tune its performance, outperforming Luma Ray 3.2 in 93% of internal evaluations.
The players
Black Forest Labs
A developer of multimodal artificial intelligence architectures focused on open-weight generative models.
mimic robotics
A Zurich-based robotics firm specializing in the development of dexterous manipulation hardware and control software.
Audi
An automotive manufacturer that serves as an industrial testing ground for robotic soft-body manipulation tasks.
The details
The architecture functions by pairing a video-prediction backbone—a system that forecasts future visual frames based on previous inputs—with a decoder that translates these sequences into specific robotic control commands. This integration of sensory modalities allows the model to handle video clips up to 20 seconds long with synchronized audio. Fine-tuning for specific operational environments is accomplished through a process called Self-Flow, which optimizes the model using only 30 minutes of physical robot demonstration data.
Timeline
September 23, 2026: Black Forest Labs launched the FLUX 3 Action model.
End of 2026: The company intends to release the weights for the FLUX 3 Dev multimodal backbone.
The Tech Race
The absence of FLUX 3 Action from the RoboLab-120 leaderboard leaves its true competitive performance against industry peers unverified. This release represents a shift toward reducing the massive data requirements that have long hindered the scaling of general-purpose robotic control.
Developers and researchers can expect to test the model's 80-millisecond latency on consumer-grade hardware for soft-body manipulation workflows. However, the model is not yet available as a finished product, and current use is limited to the initial framework release.
The takeaway
Black Forest Labs has demonstrated that fine-tuning robots on significantly reduced datasets is becoming technically feasible for complex tasks. Interested developers should monitor the official release of the FLUX 3 Dev weights scheduled for the end of 2026.
What happens next
Black Forest Labs has committed to the release of the FLUX 3 Dev multimodal backbone weights by the end of 2026.
Further reading
For more on the latest trends in hardware control and generative models, visit Robotics.
Source note: This article includes information reported by Crypto Briefing.
Live Poll
Do you trust that new AI robotic models will make local workplaces safer and more efficient?
