AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: How OlmoEarth V1.3 Creates Embeddings For The Whole Planet on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

The OlmoEarth team has released v1.3 of its satellite embedding model, annual global data at 10-meter resolution, and a browser-based explorer. The developers report that the 128-dimensional model ranked highest on 20 of 27 evaluations, but those results have not been independently verified in the source material.

The OlmoEarth team has released version 1.3 of its satellite embedding model, accompanied by annual global embeddings at 10-meter resolution and a browser-based map explorer. The release is designed to let researchers and other users analyze land cover and change using precomputed satellite features rather than first downloading and processing large imagery archives; the team’s reported performance results remain developer evaluations, not independently verified findings.

OlmoEarth v1.3 processes yearly image time series from Sentinel-1 radar and multispectral imagery from Sentinel-2 and Landsat. It produces a 128-number embedding for each 10-meter pixel, with the first 64 dimensions also available as a smaller representation. The team says the precomputed global data can be accessed through Source Cooperative.

The release also includes the OlmoEarth Embeddings Explorer, a browser-based tool for viewing embeddings as false-color maps, changing the displayed year and comparing results with Sentinel-2 imagery. According to the release description, users can select a location to find visually similar places or add labeled points and train a small classifier. The team presents the tool as usable without signing in or having advanced technical knowledge.

In its comparison with AlphaEarth Foundations and Tessera v2, the OlmoEarth team says the 128-dimensional embeddings ranked highest in 20 of 27 evaluations, while the 64-dimensional version led in 18. The evaluations covered six continents and several tasks. The team also introduced PLANTEUR, a crop-type benchmark focused on tropical farming, and says its embeddings generalized best to locations not included in classifier training. Full methods and results are in the team’s technical report.

At a glance
announcementWhen: Announced in the source material; publi…
The developmentThe OlmoEarth team released v1.3 with global annual satellite embeddings, an online explorer and developer-reported evaluation results.
At a glance
announcementWhen: Announced in the source material; the r…
The developmentThe OlmoEarth team released v1.3 alongside precomputed annual global satellite embeddings and a browser-based tool for exploring them.

Reusable Features Lower Mapping Barriers

Precomputed embeddings could reduce the storage and computing work involved in satellite mapping. Instead of repeatedly processing large image collections for each research question, users can train smaller models on the vectors or compare them across years. That may help teams without dedicated GPU resources investigate land cover and change, though the embeddings do not remove the need to validate maps or interpret them carefully.

The team says model weights, pretraining data and code are available. It estimates that reproducing the final pretraining run for OlmoEarth v1.3 Base would cost about $2,720 in rented H100 computing time. This is the developers’ estimate, not a guaranteed cost; provider prices, hardware availability and setup can change the total. Open materials may support examination and reproduction, but their availability alone does not establish how the model performs in every region or application.

The release’s reported evaluation lead is relevant to researchers choosing a reusable representation across multiple mapping tasks. Still, aggregate rankings are only one measure. A model’s usefulness for a specific project depends on local data quality, the labels available, the task being mapped and how errors affect decisions.

Products Worth Considering

Bose QuietComfort Ultra Earbuds (2nd Gen), Wireless Noise-Cancelling Earbuds, Black

Bose QuietComfort Ultra Earbuds (2nd Gen), Wireless Noise-Cancelling Earbuds, Black

Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, and IPX4 Water Resistance

Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, and IPX4 Water Resistance

Apple MacBook Air 15-inch (2026) with M5 chip, 16GB memory, 512GB SSD — Midnight

Apple MacBook Air 15-inch (2026) with M5 chip, 16GB memory, 512GB SSD — Midnight

As an affiliate, we earn on qualifying purchases.

From Task Models to Shared Embeddings

Earlier OlmoEarth workflows centered on fine-tuning a model for a particular task through OlmoEarth Studio and running it over satellite imagery to produce a map. The team says that approach requires GPU access and the ability to download and store imagery. Embeddings offer a different workflow: satellite features are processed into vectors that can then be reused for multiple classification or comparison tasks.

The v1.3 release builds on earlier versions. The team reports that training the new model used 53% of the computing time of v1.2 and 18% of the time used for v1. It also reports emissions equal to 50% of v1.2 and 22% of v1, and says v1.3 matched v1.2 in fine-tuning performance across research and partner labels. These are developer-reported comparisons; the source material does not provide the underlying emissions-accounting details.

The team illustrates possible use with maps of mining-related landscape change in Peru’s Madre de Dios region, including differences around La Pampa between 2017–2020 and 2020–2025. The release refers to Amazon Conservation’s MAAP monitoring and the 2019 Operation Mercurio crackdown. Such maps can identify changes in satellite-derived features, but an embedding difference by itself does not establish that mining caused the change.

“OlmoEarth’s 128-dimensional embeddings score highest on 20 of 27 evals and rank highest overall.”

— The OlmoEarth team

Independent Checks and Data Limits

The supplied release information does not include an independent replication of the 27 evaluations. The technical report is said to contain full methods, but assessing the results for a particular use will require examining how tasks were selected, how performance varies by region and whether the comparison applies to the intended labels and conditions.

The release description also does not specify how often the annual embeddings will be updated, what coverage or quality gaps apply in every location, or what access and computing requirements users may face for large-scale work. In the Peru example, observed differences signal change in the mapped features; attributing that change to mining, recovery or another cause requires other evidence, such as imagery, field observations or monitoring records.

The team’s reported compute and emissions comparisons also lack enough accounting detail in the supplied material to independently assess their scope. Actual reproduction costs may differ from the team’s estimate. These limits do not negate the release, but they mean users should treat performance and practical-use claims as provisional until tested against their own data and requirements.

Testing the Embeddings in Use

Researchers can access the precomputed data through Source Cooperative and try the browser explorer to inspect locations, compare years and test small classification tasks. The team’s technical report is the next source for users seeking the evaluation setup and full results; the release material does not give a date for a subsequent model update or a schedule for refreshed global embeddings.

The key next step for assessing the claims is independent testing across regions and applications, including comparisons against local reference data and checks of how well classifications generalize to places not used in training. For change maps, users will need to corroborate detected shifts with imagery or other monitoring evidence before assigning a cause. Further information about coverage, update frequency and practical access requirements would help organizations judge whether the dataset fits operational work.

Key Questions

What did OlmoEarth release?

The team released OlmoEarth v1.3, annual global satellite embeddings at 10-meter resolution and a browser-based explorer for viewing and comparing them.

What is an OlmoEarth embedding?

An embedding is a compact numerical representation learned from imagery. OlmoEarth v1.3 produces a 128-number vector for each 10-meter pixel; the first 64 dimensions can be used as a smaller representation.

How did the model perform in the team’s evaluations?

The OlmoEarth team reports that its 128-dimensional version ranked highest in 20 of 27 evaluations, while the 64-dimensional version led in 18. The supplied information does not include independent verification of those results.

Can the embeddings identify why a landscape changed?

Not by themselves. Embedding differences can indicate a change in satellite-derived features, but identifying a cause such as mining requires corroborating evidence, including imagery, monitoring records or field data.

Where can users access the data and tools?

The team says the precomputed embeddings are available through Source Cooperative and that the release includes a browser-based OlmoEarth Embeddings Explorer. The supplied material does not detail all access or large-scale computing requirements.

Primary source: Hugging Face · via ThorstenMeyerAI.com

You May Also Like

Open-source sponsor update generator

A new tool to automate sponsor updates for open-source maintainers is being tested, aiming to improve communication and support transparency.

Which Of The 10 Best AI Mini PCs For Local AI Workloads Fits Your Needs?

A new comparison ranks AI mini PCs for local workloads, led by the GEEKOM A9 Max, with the GMKtec EVO-X3 standing out at 128GB of memory.

China’s Rapid AI Release: Four Frontier-Class Open Models In Two Months

Chinese labs launched four frontier-class open models between April and June 2026, signaling rapid development in open-weight AI capabilities.

Learnings from 100K lines of Rust with AI (2025)

A developer shares insights from building a modern, high-performance Rust-based Paxos engine using AI tools, highlighting productivity, correctness, and future prospects.