Ai2 releases OlmoEarth v1.3, a satellite AI model with global embeddings and browser tool

What OlmoEarth v1.3 offers

OlmoEarth v1.3 is an AI-ready satellite embedding model that compresses a year of satellite observations into numerical fingerprints for 10-by-10-meter patches of ground. These embeddings capture patterns that distinguish land cover types such as crops, forests, and mines, enabling applications in environmental monitoring, land use classification, and change detection.

The model processes data from sources like Sentinel-1, Sentinel-2, and Landsat. Sentinel-1 measures radar reflections, while Sentinel-2 and Landsat capture visible and infrared light, providing a multi-layered view of the Earth’s surface.

Cheaper and more efficient training

OlmoEarth v1.3 reduces training costs significantly. It requires only 53% of the computing time and produces 50% of the carbon emissions compared to its predecessor, v1.2. This makes it a more sustainable and accessible option for organizations working with large-scale geospatial data.

Browser-based analysis tool

Ai2 also released OlmoEarth Embeddings Explorer, a browser-based tool that allows users to analyze satellite embeddings without technical barriers. Unlike earlier versions, which required fine-tuning in OlmoEarth Studio and access to GPU resources, the new tool enables rapid analysis of regions with minimal setup. For example, a first pass over data takes seconds, and users can classify areas using just a handful of labels without needing a GPU.

The tool democratizes access to geospatial analysis, making it feasible for researchers, NGOs, and businesses to monitor environmental changes, track land use, and identify patterns across large datasets.

Real-world application: Tracking illegal mining in Peru

OlmoEarth embeddings were used to monitor illegal gold mining in Peru’s Madre de Dios region. The model detected a peak in deforestation between 2017 and 2018, when over 1,200 hectares of forest were cleared in the buffer zone of the Tambopata National Reserve. Following Peru’s Operation Mercurio in 2019, which targeted illegal mining, the area of significant change in the southern mining belt dropped from 34 square kilometers (2016-2020) to 9 square kilometers (2020-2024).

The embeddings provided a clear, training-free way to compare land cover across years, demonstrating their utility for policymakers and environmental organizations.

How this builds on Ai2’s prior work

This release follows earlier versions of OlmoEarth, which required fine-tuning in OlmoEarth Studio and significant computational resources. OlmoEarth v1.3 improves efficiency while retaining strong performance for research and operational use. The addition of precomputed global embeddings and a browser tool removes barriers to entry, enabling faster experimentation and deployment.

The shift aligns with trends in AI infrastructure, where organizations like IBM and CoreWeave are co-designing workload controls to simplify complex tasks such as reinforcement learning and agent-based modeling. Similarly, TOTVS’s expansion of its enterprise AI foundation, LYNN, reflects a broader push to integrate AI with operational data for practical business applications.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

OlmoEarth v1.3 and its browser tool signal a shift toward accessible, GPU-free geospatial analysis. For startups and operators in agriculture, conservation, or urban planning, this lowers the cost of testing ideas—turning what was once an expensive, multi-hour process into a seconds-long query. The model’s ability to track deforestation or plantation expansion without custom training also makes it a potential alternative to proprietary geospatial platforms. Founders in Latin America or Southeast Asia, where environmental monitoring is critical, may find this tool particularly useful for compliance reporting or impact measurement.

Related on Lazyfounder

• More ai news
• IBM and CoreWeave co-design workload controls for AI agent infrastructure
• TOTVS expands enterprise AI foundation LYNN with operational data integration