MolmoAct2 open-source model achieves 87.1% success on real-world robot tasks
Allen Institute for AI and University of Washington released MolmoAct2, a fully open-source action reasoning model for robot deployment. The model achieves 87.1% success on real-world DROID tasks with unseen objects and 2.42x speedup in control rate versus unoptimized inference.
MolmoAct2, developed by the Allen Institute for AI and the University of Washington, is a fully open-source action reasoning model designed for real-world robot deployment. It achieves up to 87.1% success on real-world DROID tasks with unseen objects and a 2.42x speedup in control rate compared to unoptimized inference.
Open-source embodied AI
The model is fully open-source, reducing barriers to robotics deployment and enabling broader community iteration on agent architectures. Open availability allows researchers and practitioners to build on the same foundation without licensing restrictions.
Real-world generalization
Performance on unseen objects demonstrates generalization beyond training data. The 2.42x speedup in control rate indicates the model is optimized for practical deployment constraints. These metrics establish a baseline for open-source embodied AI systems and enable comparison with proprietary alternatives.