This product is for robotics researchers and engineers building autonomous systems for physical tasks. Today they struggle with closed-source models that lack transparency, require per-task fine-tuning, and are too slow for real-time operation. MolmoAct 2 provides an open, reasoning-based architecture that outperforms proprietary models on benchmarks, handles diverse tasks out of the box, and runs up to 37x faster than its predecessor. For example, a researcher trying to get a robot to reliably load a dishwasher or prep test tube samples in a lab can now use a single model that works without custom training and responds in near-real time.

This product is for robotics researchers and engineers building autonomous systems for physical tasks. Today they struggle with closed-source models that lack transparency, require per-task fine-tuning, and are too slow for real-time operation. MolmoAct 2 provides an open, reasoning-based architecture that outperforms proprietary models on benchmarks, handles diverse tasks out of the box, and runs up to 37x faster than its predecessor. For example, a researcher trying to get a robot to reliably load a dishwasher or prep test tube samples in a lab can now use a single model that works without custom training and responds in near-real time.