What Ai2 announced
Ai2 announced OlmoEarth v1.3 on October 22, alongside precomputed annual satellite embeddings and an Embeddings Explorer. An embedding is a set of numbers representing patterns a model has learned; here, those features describe a 10-meter grid using a year of satellite observations.
Ai2 says its browser Explorer lets people compare years, search for similar places, and make a first classification without an account or GPU. Those are the publisher’s access and workflow claims. The model repository predates today’s announcement; we are reporting the announcement and accompanying tools, not a first-ever upload of the weights.
Check the coverage and the terms
The launch archive covers land and adjacent coastal waters globally for 2023–2025. The 2017–2022 data covers selected countries, so the same historical range is not available everywhere. Ai2 also provides source-imagery provenance to help users trace the features.
The artifacts use the OlmoEarth Artifact License, which includes use restrictions and distribution conditions. Read the actual terms for your intended use; availability of weights does not mean unrestricted reuse.
Our take
For map-curious readers and teams working with land data, this is a useful example of AI beyond chat. Begin with a place and year you understand, compare the underlying imagery, and check any proposed labels against reliable local evidence.
Treat a similar-looking pattern or a changed vector as a lead to investigate. It does not, by itself, identify a crop, prove a cause, or establish current ground conditions. Annual data also needs a date beside the map. We have not run the model, tested the Explorer’s classification or export workflow, or reproduced Ai2’s benchmark claims.