IBM and NASA Release Open Lunar Foundation Model on Hugging Face

IBM and NASA have published a lunar foundation model on Hugging Face, which they describe as one of the first publicly available foundation models for scientific exploration of the Moon. The release is part of an ongoing collaboration between the two organizations to create open AI tools for scientific research. Computer Weekly reported the news.

The model is designed to help scientists analyze the massive volumes of lunar data collected over decades. According to IBM and NASA, it can surface patterns across data at a scale no single instrument could provide. The model was trained on a lunar observation dataset curated by researchers from both organizations, combining multi-instrument data to support efforts toward a sustained human presence on the Moon.

IBM and NASA suggest the model could be used to investigate several lunar phenomena, including potential ice deposits in permanently shadowed regions, irregular mare patches (volcanic features), and crater detection. Permanently shadowed regions are difficult to observe but may contain ice below the surface, which could provide water and oxygen—resources considered essential for a future Moon base and for producing rocket fuel for missions to Mars. Crater maps also help NASA select safe landing sites and plan long-term lunar infrastructure.

Alongside the model, the team has built what they call the first open source lunar dataset of its kind: a unified, machine learning-ready dataset that aggregates more than 30 spatially aligned layers from nine instruments across four missions. It combines tens of thousands of images and maps from NASA's Lunar Reconnaissance Orbiter (LRO), NASA's Grail mission, and complementary data from the Japanese Aerospace Exploration Agency's Selene mission.

The lunar model extends an established IBM-NASA collaboration. In 2024, the two organizations released a family of Earth-focused geospatial foundation models called Prithvi on Hugging Face, aimed at open science users, startups, and enterprises to simplify model training and deployment. In 2025, they released Surya, described as the first heliophysics AI foundation model trained on high-resolution solar observation data, which offers insights into the Sun's dynamic surface and can help plan for solar weather that may disrupt technology on Earth and in space.

"Uncovering the mysteries of the Moon requires an ability to learn from an extraordinary volume of scientific data," said Juan Bernabe-Moreno, director of IBM Research Europe, UK and Ireland. "The NASA-IBM Lunar Foundation Model gives scientists a foundation to explore the Moon at scale, connecting observations across instruments, revealing patterns that are difficult to see in isolation and providing an open platform the global research community can build on."

For developers and researchers, the model's availability on Hugging Face means it can be accessed and potentially fine-tuned for specific lunar science tasks. However, the source material does not specify technical requirements, licensing terms, or computational costs, so those details remain unclear. The dataset's aggregation of multiple instruments and missions may reduce preprocessing work for teams building lunar analysis tools, but the actual compatibility with existing workflows would need to be tested.

The model is positioned as an open platform, which could lower barriers for smaller research groups and startups to contribute to lunar science. Yet, as with any foundation model, its real-world usefulness will depend on how well it generalizes to new data and whether the community adopts it. IBM and NASA have not announced specific performance benchmarks or comparisons to other models.

The release follows a pattern of IBM and NASA prioritizing open science through Hugging Face, making AI tools available to a broad audience. If the lunar model follows the trajectory of Prithvi and Surya, it may see uptake in academic and commercial projects, but that remains to be seen. For now, the key takeaway is that a new, openly available resource for lunar research has entered the public domain, and its impact will depend on how the scientific and developer communities use it.