Ai2 announced that the Open Multimodal AI Infrastructure for Science (NSF OMAI) cluster is now live, marking the first milestone of a $152 million NSF-backed program to develop open AI models for scientific research.

MolmoPoint unveiled a token-based grounding mechanism for spatial reasoning that achieves state-of-the-art accuracy, while Olmo Hybrid pairs transformer attention with linear RNNs to match prior models using roughly half as much training data.

NSF OMAI emphasizes open, shared compute to foster reproducibility and long-term value beyond a single commercial product, countering trends of concentrated access to advanced AI systems.

The cluster is deployed and managed by Cirrascale Cloud Services, reinforcing Ai2’s existing partnership for distributing Olmo, Molmo, and Tulu AI models.

Ai2 leads OMAI with collaboration from principal investigators at the University of Hawaii at Hilo, the University of New Hampshire, the University of New Mexico, and the University of Washington, aligning with the White House AI Action Plan.

NSF Deputy Director Wendy Nilsen emphasized that open, shared AI resources accelerate discovery, improve scientific rigor through replicability and transparency, and bolster U.S. leadership in AI research.

The OMAI project, supported by the National Science Foundation and NVIDIA, aims to create a fully open AI ecosystem to accelerate discovery in materials science, biology, and energy, with Cirrascale Cloud Services managing the infrastructure.

Overall, the initiative is framed as an open-science effort to democratize access to AI tooling and artifacts to advance cross-disciplinary scientific research.

Molmo 2 demonstrated the cluster’s impact by adding video understanding, pointing, and object tracking with an 8B-parameter model that outperformed the original 72B Molmo on key benchmarks.