NASA and IBM release an open AI model for lunar research

NASA and IBM Research have released a foundation model designed for lunar science. NASA says it is publicly hosted on Hugging Face, with code available on GitHub for research and experimentation.

The model was trained primarily with Lunar Reconnaissance Orbiter observations, using roughly two million image tiles. Scientists can adapt it to tasks such as identifying craters, mapping unusual volcanic terrain and estimating locations where polar ice might remain stable.

NASA describes the tool as a way to make a large scientific archive more usable. Its outputs support analysis; an estimate of ice stability is not the same as confirming that accessible ice exists at a particular location.