OlmoEarth v1.3 opens planet-scale satellite embeddings
Ai2 released OlmoEarth v1.3, an open satellite foundation model that compresses annual Sentinel-1, Sentinel-2, and Landsat imagery into 128-dimensional embeddings for each 10-meter land pixel. Precomputed global data and a no-login browser explorer make crop mapping, land-cover analysis, and change detection far more accessible.
OlmoEarth v1.3 turns expensive geospatial modeling into a lightweight feature-engineering problem, making serious Earth-observation AI practical for smaller teams.
- –Precomputed embeddings eliminate the need to download massive imagery archives or run the foundation model locally.
- –Ai2 reports top performance on 20 of 27 evaluations against AlphaEarth Foundations and Tessera v2.
- –The 64-dimensional variant offers a cheaper path for large-scale classifiers and similarity searches.
- –Open weights, data, code, provenance, and browser tooling create a strong foundation for reproducible AI-for-science workflows.
- –The main limitation is that downstream users still need labeled data and geospatial expertise to build reliable task-specific maps.
DISCOVERED
1h ago
2026-10-23
PUBLISHED
1h ago
2026-10-23
RELEVANCE
AUTHOR
CalatheaAI