- Oct 12, 2026 Introducing Reflex – within 1 s, a humanoid sees a flying box, anticipates its motion, and reacts with whole-body reflexes to catch it. Near-privileged-state performance in simulation (85.7% vs. 87.9%); real G1 demos with throws from 2.5–7 m. Check out the paper!
- Sep 24, 2026 Four papers accepted to NeurIPS 2026: MolmoMotion (Spotlight), RefDecoder, VFIG, and FramePack-U!
- Sep 04, 2026 MolmoAct2 – an open foundation for robots that work in the real world – is accepted to CoRL 2026! Check out the tech report, data, pre-trained models, code.
- Sep 04, 2026 TOPReward – a training-free reward model for robotics – is accepted to CoRL 2026!
- Aug 16, 2026 I’m teaching COMP 690: Deep Generative Models at UNC-Chapel Hill this Fall 2026, covering the theory and applications behind modern image, video, 3D, action, and language generation systems. If you’re a UNC student interested in generative modeling, please register!
- Jun 17, 2026 Introducing MolmoMotion – 3D point trajectories forecasting model from language instruction! We are excited to release the paper, data, benchmark, pre-trained models, code.
- Jun 17, 2026 MolmoWeb – open visual web agent for automating browser tasks – is accepted to ECCV 2026! Check out the paper, data, pre-trained models, code, demo.
- Jun 07, 2026 RefDecoder received the 🏆 Best Paper Award at the AI for Content Creation (AICC) Workshop @ CVPR 2026!
- Apr 08, 2026 Molmo2 has been accepted into CVPR 2026 as oral presentation (top 0.7% of submissions) and received the 🏆 Best Paper Award Candidate and 🏆 CVPR Compute Transparency Champion Award! We’ve released everything: tech report, online demo, full data, pre-trained models, and code.
- Apr 07, 2026 Introducing WildDet3D – open 3D detector for objects in-the-wild! We are excited to release the tech report, online demo, data, benchmark, pre-trained models, code, and Iphone App.
- Mar 28, 2026 Thrilled to introduce VFig: a powerful Img-to-SVG VLM and MolmoWeb: an open agent for automating web tasks.