Releases·Model Releasefeatured
BFL releases FLUX 3 Action, a 7B open-weights World Action Model topping RoboLab
Black Forest Labs released FLUX 3 Action, an open-weights 7B World Action Model that predicts video and actions jointly for robot control. It ranks first on the RoboLab benchmark, beating the best open model by 6.1 points while running up to 3.95x faster. Teams can fine-tune it on their own demonstrations, and BFL reports early results beyond robotics in gaming, vehicle control, and computer use.
145 engagement2 sourcesWed, Sep 23, 2026, 05:56 PM
AI Overview
- Black Forest Labs released FLUX 3 Action, a 7B-parameter model for robots, per bfl.ai.
- The model page is live at bfl.ai/models/flux-3-action, linked via a Hugging Face post.
- The announcement surfaced on Reddit on 2026-09-23 with modest engagement (29).
- FLUX 3 Action ranks first on the RoboLab benchmark, beating the prior best open model by 6.1 points with 56% fewer parameters and up to 3.95x faster speed.
- The model predicts video and actions jointly, plans over 2x further ahead, and runs faster per second of robot motion than the strongest open VLA.
- Built on FLUX 3's image, video, and audio pretraining with a smaller deployment-focused architecture, midtrained to predict actions and future frames together.
- BFL reports training task-specific policies beyond robotics: gaming, vehicle control, computer use, and other simulated visual environments.
- Teams can fine-tune FLUX 3 Action on their own demonstrations for specific robots and tasks; weights are on Hugging Face.
Related
Releases·Model Releasefeatured
Black Forest Labs releases FLUX 3, a unified multimodal model for image, video, audio, and action prediction
Black Forest Labs (BFL) announced FLUX 3, a single multimodal flow model that handles image, video, audio, and action prediction. The model is positioned as a step toward a visual intelligence backbone, though specific parameter counts, pricing, and benchmark numbers were not disclosed. The release was noted as a major AI launch on July 23-24, 2026.
Black Forest Labs (BFL) announced FLUX 3, a single multimodal flow model that handles image, video, audio, and action prediction. The model is positioned as a step toward a visual intelligence backbone, though specific parameter counts, pricing, and benchmark numbers were not disclosed. The release was noted as a major AI launch on July 23-24, 2026.
140345 engagement5 sourcesreddit, rssThu, Jul 23, 2026, 05:39 PM
Releases·Model Release
Xiaomi releases XR-1 robotics foundation model trained on 100K hours
Xiaomi released Xiaomi-Robotics-1 (XR-1), a Vision-Language-Action (VLA) robot foundation model trained on over 100K hours of real-world manipulation trajectories. It uses a two-stage training paradigm—pre-training for breadth and post-training for alignment—to enable out-of-the-box mobile manipulation in unseen environments and efficient adaptation to new tasks.
Xiaomi released Xiaomi-Robotics-1 (XR-1), a Vision-Language-Action (VLA) robot foundation model trained on over 100K hours of real-world manipulation trajectories. It uses a two-stage training paradigm—pre-training for breadth and post-training for alignment—to enable out-of-the-box mobile manipulation in unseen environments and efficient adaptation to new tasks.
7 engagement1 sourceredditWed, Aug 5, 2026, 08:27 PM
Developments·Paper
Lumo-2 latent world-action model introduced for robot learning
Researchers introduced Lumo-2, a latent world-action model that generates actions by reasoning over world dynamics in latent space. It captures physically grounded visual transitions for predictive reasoning and cross-modal alignment, aiming to improve robot learning scalability and alignment.
Researchers introduced Lumo-2, a latent world-action model that generates actions by reasoning over world dynamics in latent space. It captures physically grounded visual transitions for predictive reasoning and cross-modal alignment, aiming to improve robot learning scalability and alignment.
0 engagement1 sourcearxivMon, Jul 13, 2026, 08:53 AM
Releases·Model Release
OpenBMB open-sources MiniCPM-Robot embodied AI model series with 1.5B manipulation and 0.5B tracking models
OpenBMB released MiniCPM-Robot, an open-source embodied AI model series including MiniCPM-RobotManip (1.5B parameters) for robotic manipulation and MiniCPM-RobotTrack (0.5B parameters) for target tracking, along with the PhyAI inference framework. The models are designed to enable robots to understand, remember, and act in the physical world.
OpenBMB released MiniCPM-Robot, an open-source embodied AI model series including MiniCPM-RobotManip (1.5B parameters) for robotic manipulation and MiniCPM-RobotTrack (0.5B parameters) for target tracking, along with the PhyAI inference framework. The models are designed to enable robots to understand, remember, and act in the physical world.
69 engagement1 sourceredditMon, Jul 20, 2026, 09:14 AM
Releases·Model Release
NVIDIA releases Cosmos 3 Edge, a 4B-parameter open world model for edge AI
NVIDIA released Cosmos 3 Edge, a 4-billion-parameter open world model designed for edge devices in factories, warehouses, and hospitals. The model enables robots and vision AI agents to understand surroundings, reason in real time, and generate actions on memory-constrained systems. It is available on Hugging Face's Cosmos 3 repository.
NVIDIA released Cosmos 3 Edge, a 4-billion-parameter open world model designed for edge devices in factories, warehouses, and hospitals. The model enables robots and vision AI agents to understand surroundings, reason in real time, and generate actions on memory-constrained systems. It is available on Hugging Face's Cosmos 3 repository.
0 engagement1 sourcerssMon, Jul 20, 2026, 03:58 PM