A More Efficient Family of Models: Introducing OlmoEarth v1.1
OlmoEarth v1.1: A New Generation of Remote Sensing Models
We recently released OlmoEarth (v1) in November 2025, and since then, our partners have been applying 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 just days. Every release moves us closer to our goal: making state-of-the-art AI accessible for organizations and communities working towards protecting people and the planet.
For OlmoEarth, efficiency is key, especially when it comes to processing satellite imagery over large areas. The compute cost of running the model scales quadratically with the sequence length of tokens it processes. Reducing this length by even a small amount can make a significant difference in deployment costs.
Key Takeaways
- OlmoEarth v1.1 cuts compute costs by up to 3x while maintaining performance on key tasks and benchmarks.
- The model family uses fewer tokens per resolution, leading to significant savings in pretraining, fine-tuning, and inference phases.
- This approach allows for more efficient use of resources without sacrificing the model’s ability to capture important cross-band relationships in Sentinel-2 imagery.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
