Overview
- NASA and IBM Research jointly published the NASA‑IBM Lunar Foundation Model and its SomBench benchmark as open‑source tools for researchers and the public.
- The tools were trained on nearly 20 years of orbital data, led by the Lunar Reconnaissance Orbiter and supplemented by missions such as GRAIL and SELENE.
- SomBench assembles roughly two million aligned image tiles and illumination metadata so models can handle extreme lunar lighting and surface angles.
- Benchmark results released by the teams show about 22–23% accuracy gains on selected lunar‑mapping tasks compared with common baselines.
- NASA and IBM say the model is a research accelerant to help find landing sites and likely ice deposits but stress that outputs need independent validation and scientist review before any mission use.