Ai2 releases OlmoEarth v1.3 with reusable 10-meter embeddings for the planet
The 128M-parameter remote-sensing model ships with 64- and 128-dimensional annual embeddings, while its custom license limits some downstream uses.
2 min read
Ai2 has released OlmoEarth v1.3, a 128-million-parameter model that turns satellite observations into reusable geographic embeddings. The release includes annual, global representations for 2023–2025 at 10-meter resolution in both 64- and 128-dimensional formats, plus an interactive browser for exploring locations without first running the model.
OlmoEarth combines Sentinel-1 radar, Sentinel-2 optical imagery and Landsat data. Instead of asking users to process the raw imagery for every downstream task, Ai2 distributes compact vectors that can feed classifiers and retrieval systems for applications such as crop mapping, land-cover monitoring and disaster assessment. Earlier years are available for selected countries rather than globally.
Ai2 reports that the 128-dimensional release led AlphaEarth Foundations and Tessera on 10 of 11 directly comparable tasks in its k-nearest-neighbor evaluation; the 64-dimensional version led on seven. Those are publisher-run results, however, not an independent audit. A k-NN probe also measures the usefulness of frozen embeddings under a particular evaluation recipe, not end-to-end performance for every remote-sensing workflow.
The canonical training repository, maintained by AllenAI, had 323 total GitHub stars and 50 forks when checked October 22, 2026. That is a point-in-time popularity measure; no historical snapshot was available to support a claim about recent star growth. The repository provides training, inference and evaluation code.
Access also comes with an important qualification: OlmoEarth uses a custom Artifact License restricting military, surveillance, policing and some extractive uses, so it should not be described as conventionally open source. Annual 10-meter embeddings are useful for broad monitoring, but they are not real-time imagery and may miss fine-grained or rapidly changing conditions.
Sources: Ai2 announcement · Project page · Canonical repository
Conversation
0 approved commentsNo comments yet. Start a thoughtful conversation.
Join the conversation.
Sign in with ChatGPT to leave a comment or rate this article.