MolmoAct 2 is an open robotics foundation model (a pre-trained AI system that robots use to understand and act in the physical world) built by the Allen Institute for AI.
Deploy robots on new tasks out of the box — no per-task fine-tuning required
Run inference (the time it takes a robot to decide its next move) in as little as 180ms, enabling near-real-time robot responses
Access the largest open-source bimanual (two-arm) robotics dataset ever released, with over 720 hours of training demonstrations
Use built-in 3D spatial reasoning to help robots understand their environment before acting
Download fully open model weights, datasets, and training code for research and development
Source: allenai.org
No integrations listed yet for MolmoAct 2.
MolmoAct 2 uses a specialized embodied-reasoning AI backbone (Molmo 2-ER) trained on 3 million examples to help robots reason about 3D space, detect objects, and plan physical actions in near-real time.
MolmoAct 2 is built for robotics researchers and AI engineers, not small business owners. If you run a warehouse, lab, or manufacturing operation and have a dedicated robotics team, this open model gives you a powerful, free starting point. But if you don't already have robotics hardware and machine learning engineers on staff, you won't be able to use this tool — it's not a plug-and-play product. Wait for commercial products built on top of models like this before considering it for your business.
AI-generated training guides tailored to your team's size, skill level, and focus areas for MolmoAct 2 — coming in v0.3.2.
View our roadmap →We're building a review system so business owners like you can share real experiences with MolmoAct 2.
Last researched: May 2026
