IBM and NASA Open-Source Lunar AI Model for Moon Exploration

IBM and NASA have released the NASA-IBM Lunar Foundation Model, an open-source AI system designed to help scientists analyze decades of lunar observation data and accelerate discovery on the Moon's surface. The model, announced September 10, 2026, represents one of the first publicly available foundation models specifically built for lunar scientific exploration.

The system was trained on an extensive dataset curating information from nine instruments across four missions, including NASA's Lunar Reconnaissance Orbiter, NASA's GRAIL mission, and the Japanese Aerospace Exploration Agency's SELENE/Kaguya spacecraft. This unified dataset aggregates over 30 spatially-aligned data layers combining tens of thousands of images and maps that capture unique geophysical properties of the lunar surface and subsurface.

Performance gains across key lunar features

The model demonstrates measurable improvements over widely used computer vision methods in identifying critical geographic features. For detecting potential lunar ice deposits in permanently shadowed regions, the NASA-IBM model reduced error by up to 22% compared to the SwinV2-B ImageNet model, according to a technical paper authored by IBM and NASA researchers.

In crater detection at context-scale resolution (approximately 100 meters), the model outperforms SwinV2-B by nearly 19% while using only half the training data. For mapping Irregular Mare Patches—volcanic features that reveal the Moon's thermal evolution—the model captures the extent of these formations 3% better than existing approaches while requiring lower fine-tuning costs.

Why it matters

Identifying lunar ice deposits is essential for establishing a sustained human presence on the Moon, as ice indicates water and oxygen resources needed for a future Moon base and producing rocket fuel for Mars missions. The model's ability to synthesize multi-modal, multi-resolution data addresses a longstanding challenge: scientists previously had to manually examine maps and images or rely on low-resolution, task-specific machine learning models that lacked the accuracy needed for precise geographic analysis. By providing a unified foundation that connects observations across instruments, the model enables researchers to surface patterns at a scale no single instrument could provide.

Open foundation for the research community

Alongside the model, IBM and NASA built the first open-source, machine-learning-ready lunar dataset that brings multi-modal, multi-resolution data into a common framework. No such publicly available unified dataset previously existed for the Moon.

The release extends IBM and NASA's collaboration on the Prithvi family of open foundation models, which now spans geospatial, weather, heliophysics, and lunar domains. Rather than building separate algorithmic systems for every scientific question, researchers can adapt this shared model to new tasks.

"NASA has spent decades building an extraordinary scientific record of the Moon, but collecting data is only part of the job," said Kevin Murphy, chief science data officer and acting chief data and AI officer at NASA Headquarters. "We also have to make data easier for scientists to explore and use."

Juan Bernabe-Moreno, Director of IBM Research Europe, UK and Ireland, noted the model "gives scientists a foundation to explore the Moon at scale, connecting observations across instruments, revealing patterns that are difficult to see in isolation."

The details were first reported by IBM in a September 10, 2026 announcement from Yorktown Heights, New York.