New Satellite AI Model Simplifies Global Mapping
Ai2’s OlmoEarth v1.3 provides precomputed satellite vectors, enabling researchers to skip resource-heavy training tasks.
Updated on Oct. 23, 2026 in Artificial Intelligence
Live Poll
Do you believe developing open-source AI models for global mapping benefits the public?
The Allen Institute for Artificial Intelligence released OlmoEarth v1.3, an open-source model providing satellite data at 10-meter resolution. The model offers annual embeddings for 2023 through 2025 across most of the globe, excluding Antarctica.
Why it matters
By providing precomputed annual embeddings for satellite imagery, this release allows researchers to perform large-scale mapping and change analysis without the computational cost of rerunning training models.
OlmoEarth v1.3 processes imagery into 64- or 128-dimensional vectors for every 10-meter pixel. The model required 539 H100 GPU hours for pretraining, resulting in a carbon footprint of 0.07 metric tons of CO2 equivalent.
The players
Allen Institute for Artificial Intelligence
A research organization focused on open-source machine learning, known for developing transparent, accessible AI models for scientific applications.
The details
The model functions by transforming raw satellite input from Sentinel-1, Sentinel-2, and Landsat into standardized vector representations. This vectorization allows users to perform similarity searches and clustering on geospatial data through a web-based explorer. The platform further supports integration into external workflows by enabling the export of these data representations via Python scripts.
Timeline
2017-2022: Limited coverage for selected countries.
2023-2025: Annual global embeddings available.
October 22, 2026: Ai2 released OlmoEarth v1.3.
The Tech Race
This release mirrors the Allen Institute for Artificial Intelligence's previous efforts to open-source foundation models like OLMo to accelerate domain-specific discovery. It positions the institute as a key supplier of open-access infrastructure for global geospatial analysis.
Researchers can immediately access the precomputed embeddings through the project's browser-based explorer or export data for local use in Python. The release removes the need for individual teams to spend hundreds of GPU hours processing raw satellite data.
The takeaway
OlmoEarth v1.3 provides a significant efficiency gain for environmental monitoring by shifting the burden of expensive, large-scale satellite feature extraction to a single, shared open-source baseline. Researchers should monitor future updates from the Ai2 GitHub for expanded temporal coverage beyond 2025.
Further reading
For more developments on foundational modeling and open-source research, visit Artificial Intelligence.
Source note: This article includes information reported by TokenPost.
Live Poll
Do you believe developing open-source AI models for global mapping benefits the public?
