An open AI model helps scientists map the Moon

In September 2026, NASA released a new artificial intelligence model that is designed to help scientists study the surface of the Moon. The space agency announced it in an article published on September 10. Known as the Lunar Foundation Model, it was built by NASA together with IBM Research and researchers from several universities. The model is open, which means that anyone can use it, and its complete computer code has been shared online.

So what exactly is a foundation model? Traditionally, if scientists wanted a computer to do one job, such as finding craters, they had to build and train a special program for that single task, starting from nothing. A foundation model works in a different way. First, it is trained on a huge amount of data that nobody has labeled, so it learns general patterns by itself. This stage is called pre-training. Researchers can then adapt the model to a new task by using only a small number of labeled examples, such as images in which experts have already marked every crater. This quicker second step is known as fine-tuning.