OlmoEarth v1.1: A more efficient family of models

Update 6/29: OlmoEarth v1.2 now available

Today we're releasing OlmoEarth v1.2, which implements rotary positional embeddings (RoPE)—reducing artifacts in the embeddings and giving a small performance boost. This update came directly from partners asking for cleaner embeddings. 

OlmoEarth processes satellite images into tiles (patches) and turns each into an embedding, a numerical representation. Older versions marked patches’ positions with a fixed signal (absolute positional embeddings, or APE) and that signal showed up as extraneous artifacts in the embeddings. RoPE fixes this: Unlike APE, RoPE rotates vectors the model compares according to their distance. This encodes patch location without introducing artifacts in the embeddings.

Across all model sizes, we see consistent improvement on our kNN/linear-prob evals.

OlmoEarth v1.2 comes in Nano, Tiny, Small, and Base—all open source and available now. If you're already using Base, we recommend that you keep using it, but if you're on Tiny or Nano, we suggest moving to Small. For more information, read our technical report. 

We released OlmoEarth (v1) in November 2025. Since then, partners have applied it across a wide range of tasks, from tracking mangrove change to classifying drivers of forest loss to producing country-scale crop-type maps in days, scaling deployments to national, continental, and global areas. Every release moves us closer to our mission: bringing state-of-the-art AI to organizations and communities working to protect people and our planet.