NSF OMAI goes live: Ai2's fully open AI cluster is a bet against closed research

The Allen Institute for AI (Ai2) announced today that the Open Multimodal AI Infrastructure for Science (NSF OMAI) has shifted from blueprints to reality. The project, backed by $152 million from the U.S. National Science Foundation and NVIDIA, now runs on a cluster built around NVIDIA Blackwell Ultra systems, designed to nurture a fully open AI research ecosystem.

“At a time when access to advanced AI systems is increasingly concentrated among a small number of companies, bringing this hardware infrastructure online represents a critical step for us,” said Noah A. Smith, Principal Investigator for NSF OMAI and Senior Research Director at Ai2. “Our goal is to accelerate a truly open technology ecosystem with broad impact.”

From infrastructure to impact

In closed systems, substantial compute is spent on experiments and intermediate results that rarely leave the organization, often yielding only a single final product for commercial use. If that same compute generates an open artifact, it continues to generate value long after training ends. Data, checkpoints, methods, and final models can be picked up and adapted across many downstream applications, allowing other labs to avoid repeating costly experiments.

Recent internal research from Ai2 estimates that, in some cases, 82% of training effort goes toward exploratory work rather than the final model. When shared, each GPU hour contributes not just to one release but to a growing body of work the entire field can draw on. The result is a multiplier effect: the same resources support more ideas, applications, and progress over time, without the footprint required by large, closed deployments.

The new cluster reflects this philosophy. Built on NVIDIA B300 systems, it prioritizes how effectively capacity is used and shared, rather than relying on raw scale. Deployed and managed in partnership with Cirrascale Cloud Services, it supports both large-scale training and ongoing experimentation across language, multimodal, and scientific domains.

“NSF OMAI reflects our commitment to ensuring that advanced AI infrastructure supports the broader research community,” said Wendy Nilsen, Deputy Directorate Head of the NSF Computer and Information Science and Engineering Directorate. “By investing in open, shared resources, we are enabling scientists and researchers to build, test, reproduce, and advance AI systems.”

Jack Wells, Director of Higher Education and Research Computing at NVIDIA, added: “By executing the building of the NSF OMAI project cluster on NVIDIA Blackwell Ultra, Ai2 is creating a highly efficient, open ecosystem that maximizes the impact of every compute hour.”

Looking ahead: powering the future of open source AI

As the cluster comes online, Ai2's work across language and multimodal disciplines is converging, reflecting a greater focus on unified architectures that handle multiple data types and tasks natively. Ai2 is also continuing to invest in models that act as agents, ones that can plan, use tools, and act autonomously in complex environments. Some of this work has already been published, like the Open Coding Agents family, MolmoWeb, and ongoing research into how training strategies shape reliable agentic behavior.

Alongside this, Ai2 is improving its infrastructure for training and evaluation to ensure the systems used to build and benchmark models can scale with the research. The Olmo team's researchers are also conducting outreach to science communities to make sure the next generations of models are genuinely useful for those fields.

“As a member of the original Olmo team, I'm excited to be back at Ai2 at this pivotal moment as we continue to advance fully open AI for science,” said Iz Beltagy, Research Lead of the Olmo Team at Ai2.

Research already yielding results

Research supported through NSF OMAI is already producing concrete outputs:

• Molmo 2 introduced video understanding, pointing, and object tracking to Ai2's multimodal model family, with an 8B-parameter model surpassing the original 72B Molmo on key benchmarks. Nine new datasets covering tasks such as advanced video grounding and multi-image grounding were released under a permissive license.
• MolmoPoint followed with a new pointing architecture replacing text-coordinate outputs with a token-based grounding mechanism, achieving state-of-the-art accuracy on spatial reasoning tasks.
• Olmo Hybrid combined transformer attention with linear RNN layers in a new architecture that matches prior models while using roughly two times less training data.
• Work on meta-reinforcement learning with self-reflection is advancing how search agents learn from prior attempts, improving exploration without external reward models.

These projects illustrate the breadth of research NSF OMAI is accelerating across Ai2's language modeling programs, producing not just models, but open artifacts that other teams can inspect, adapt, and build on.

“Olmo uniquely enabled [my research] work because it is fully open source,” said Yuan He, a former research associate at the University of Oxford. “In a landscape dominated by commercially optimized systems, Olmo stands out by empowering deeper understanding rather than just application.”

About NSF OMAI

In 2025, Ai2 was awarded a cooperative agreement through the NSF's Mid-Scale Research Infrastructure program, combined with a $77 million investment from NVIDIA, to form NSF OMAI. The project aims to empower the AI community to inspect, reproduce, and innovate, while transforming scientific discovery across disciplines. Ai2 leads alongside co-PIs from the University of Hawai'i Hilo, the University of New Hampshire, the University of New Mexico, and the University of Washington.