OlmoEarth: The Open-Source AI That's Giving Us New Eyes on the Planet

We are drowning in data, yet starving for wisdom.

This paradox is most apparent when we look at our own planet. Every day, a fleet of satellites—both public and private—capture petabytes of information. They see our world in spectrums invisible to the human eye, tracking changes in our forests, oceans, cities, and ice caps with relentless precision. We have more raw data about Earth than ever before.

But here’s the problem: data is not insight. A pixel is not a policy. A terabyte of signals is not an action plan.

For decades, the challenge for scientists, governments, and NGOs has been bridging the vast chasm between raw satellite signals and actionable intelligence. How do you turn a specific spectral signature over the Amazon into a real-time deforestation alert? How do you translate soil moisture readings across a continent into a reliable wildfire risk forecast?

This translation process has been slow, expensive, and fragmented. It required bespoke, single-purpose AI models that were often proprietary, difficult to build, and impossible to share or adapt.

Until now.

The Allen Institute for AI (AI2) has just introduced OlmoEarth, a new family of open-source AI foundation models built specifically for Earth Observation (EO). This isn’t just another language model or a new photo app. OlmoEarth is a fundamental shift in how we use AI to understand our planet. It’s a tool designed to finally turn that flood of data into a stream of wisdom. And it’s open for everyone.

This is the story of OlmoEarth, what it is, what it can do, and why it might be one of the most important AI releases for the future of our planet.

What Exactly is OlmoEarth? The Foundation for a Smarter Planet

First, let’s be clear about what OlmoEarth is not. It is not a chatbot. It won’t write you a poem or a marketing email. Its “language” isn’t English or Python; its language is the complex dialect of multispectral satellite imagery.

OlmoEarth is a foundation model. This is a critical concept.

Think of a foundation model like a university graduate who has just completed a rigorous, multi-disciplinary degree in “Earth Science.” They understand the fundamentals—what a river looks like from 400 miles up, how different crops reflect infrared light, the seasonal patterns of snowmelt. They have a deep, general-purpose understanding of the subject.

Because they have this strong foundation, you can hire them and, with a little specific training (called “fine-tuning”), they can become a world-class expert in a niche task. You can train them to become a “mangrove health specialist” or a “wildfire fuel-load analyst.”

This is the power of the OlmoEarth models. AI2 has already done the heavy lifting of pre-training OlmoEarth on vast, diverse, global satellite data. They have built the “graduate.” Now, researchers, communities, and organizations can take this powerful base model and quickly fine-tune it for the specific, local problems they need to solve.

The OlmoEarth project is being released as a family of models, offering different sizes for different needs, from the lightweight “Nano” and “Tiny” models to the more powerful “Base” and “Large” versions. The OlmoEarth-v1-Base model, with 89 million parameters, is presented as the ideal starting point for most users, balancing power with accessibility.

The “open” in “open-source” is the other half of the revolution. By making OlmoEarth completely open, AI2 is democratizing access to this high-end technology. Governments in the developing world, under-funded conservation groups, and local community activists can now access the same cutting-edge AI tools as major corporations and military agencies. This levels the playing field in the fight against climate change.

The OlmoEarth Ecosystem: More Than Just a Model

AI2 understands that a powerful model is useless if it’s too hard to use. A brilliant AI sitting on a server helps no one. That’s why they didn’t just release code; they released an entire ecosystem designed to accelerate adoption and make OlmoEarth practical from day one.

This ecosystem consists of three key parts:

1. The OlmoEarth Platform and Studio

This is the main entry point for most teams. The OlmoEarth Platform is an end-to-end solution designed to be the “fastest, most cost-effective way to get from global satellite data to real-time Earth insights”.

The star of this platform is the OlmoEarth Studio, a free, web-based workspace. This is a game-changer. The Studio allows teams to:

• Create Datasets: Upload your own data or access existing public data.
• Fine-Tune Models: Use the intuitive interface to train a base OlmoEarth model for your specific task, like “identify new construction” or “map crop disease”.
• Collaborate: Work as a team to build, iterate, and perfect your custom OlmoEarth model.

The OlmoEarth Studio effectively removes the barrier of complex infrastructure. You don’t need to be a DevOps expert or have a cluster of GPUs to start using OlmoEarth.

2. The Hugging Face Collection

For the developers and data scientists who want to get their hands dirty, the entire OlmoEarth model family is hosted on Hugging Face, the central repository for the AI community.

This collection (allenai/olmoearth) is the “go-bag” for OlmoEarth. It includes:

• The Base Models: The pre-trained, general-purpose OlmoEarth models (Base, Tiny, etc.) ready for download.
• Fine-Tuned Models: A growing library of ready-to-use models that are already specialized for specific tasks. This is incredibly valuable.

Want to map all the mangrove forests in a region? There’s an OlmoEarth fine-tuned model for that. Need to classify different types of ecosystems? There’s an OlmoEarth model for that, too. This collection is a testament to the “build and share” philosophy of the OlmoEarth project.

3. The GitHub Code Repositories

This is the “engine room” of OlmoEarth. AI2 has open-sourced the code in two main repositories:

• olmoearth_pretrain: This is the core DNA. It’s the code used to create the OlmoEarth foundation models themselves. Built on the OLMo-core framework, this is for the advanced researchers who want to understand how OlmoEarth was built, replicate it, or even build their own foundation models from scratch.
• olmoearth_projects: This is arguably the most valuable repository for new users. It’s a collection of real-world projects and examples showing how to use OlmoEarth models for specific, practical tasks.

This three-pronged approach—an easy-access Studio, a community-driven model hub, and fully transparent code—is what makes OlmoEarth a true platform, not just a passing project.

What Can OlmoEarth Actually Do? (From Pixels to Action)

This is the most exciting part. The abstract potential of OlmoEarth is already being translated into concrete, real-world applications. The fine-tuned models and project examples show a clear path from data to decision-making.

Here are just a few examples of what OlmoEarth is already being trained to do:

1. Land Cover and Ecosystem Mapping

Understanding what is on the ground is the first step to protecting it. OlmoEarth excels at this.

• Detailed Land Use: There’s already a fine-tuned OlmoEarth model specifically for land cover mapping in Kenya. This model can automatically distinguish between farmland, urban areas, forests, and water bodies from a satellite image.
• Vital Ecosystems: The community is building models to map critical, high-value ecosystems. The OlmoEarth mangrove mapping model, for example, can identify these vital coastal forests, which are essential for carbon storage and storm protection. Another model focuses just on mapping diverse ecosystem types.

2. Predictive Disaster Management

This is where OlmoEarth moves from reactive mapping to proactive warning. The olmoearth_projects repository shows a project for mapping “Live Fuel Moisture Content” (LFMC).

This sounds technical, but its implication is profound. LFMC is a measure of how much water is in living vegetation, like grasses and shrubs. It is one of the single most important factors in predicting wildfire risk. An OlmoEarth model that can accurately map LFMC in near-real-time across a vast, remote landscape is an incredibly powerful tool for fire departments. It allows them to pre-position resources, issue targeted warnings, and manage controlled burns. This OlmoEarth application could, without exaggeration, save lives and property.

3. Investigating Environmental Change

The OlmoEarth platform isn’t just about what’s happening, but why it’s happening. Another project example is a classifier for “Forest Loss Drivers”.

When a patch of forest disappears, this OlmoEarth model can help answer the “why.” Was it a wildfire? Was it commercial logging? Was it conversion to agriculture? Knowing the driver of deforestation is essential for policymakers. You can’t fight illegal logging with a wildfire prevention plan. This specialized OlmoEarth model provides the granular insight needed to create effective environmental policy.

These are just the first examples. Because OlmoEarth is an open foundation, anyone can come up with a new application. Imagine OlmoEarth models fine-tuned to:

• Monitor illegal fishing vessels by identifying their unique wakes.
• Track the spread of invasive species in farmland.
• Estimate rooftop solar panel installations to map renewable energy adoption.
• Detect plastic-waste aggregations in the ocean.

The potential of the OlmoEarth platform is limited only by the community’s imagination.

Getting Started with OlmoEarth: Your Role in the Ecosystem

So, how do you get involved? Whether you’re a developer, a data scientist, a researcher, or just part of an organization that needs environmental insights, the OlmoEarth ecosystem has a path for you.

For Organizations and Research Teams:
Your best starting point is the OlmoEarth Studio. You can sign up for the free workspace and begin experimenting with your own data immediately. This is the fastest, most direct path to building a custom-tuned OlmoEarth model for your specific problem.

For Developers and Data Scientists:
You’ll want to head straight to the OlmoEarth Hugging Face collection and the GitHub project repo.

1. Pull down the OlmoEarth-v1-Base model as your sandbox.
2. Explore the olmoearth_projects repository to see how to structure a fine-tuning task.
3. Find a public dataset (or create your own) and try to fine-tune your first specialized OlmoEarth model.

Working with powerful foundation models like OlmoEarth is exciting, but it often requires a robust stack of tools for experimentation, deployment, and scaling. While OlmoEarth Studio provides a great starting point, many developers and teams want to integrate models like OlmoEarth into their own custom AI pipelines, web applications, or data-science workflows.

This is where having a reliable, high-performance platform for AI development becomes critical. For instance, many top developers now rely on integrated AI toolchains to manage their workflows, from data-prep to inference.

If you’re building or scaling your own AI applications, whether they’re for Earth Observation with OlmoEarth or for language processing, you need the right tools. We’ve seen platforms like ray3.run emerge as essential hubs for developers looking to build, deploy, and share powerful AI creations efficiently. Having a solid workbench is half the battle, and these platforms provide the infrastructure that lets you focus on the creative part: building something new.

For Everyone:
The OlmoEarth team is actively seeking feedback. They have a dedicated OlmoEarth-Feedback repository on GitHub for bug reports and feature requests for the platform and Studio. This open-door policy is rare and shows a real commitment to building OlmoEarth with the community, not just for it.

The Future is Open: Why OlmoEarth Truly Matters

It’s easy to get lost in the technical details, but it’s worth zooming out to understand the long-term significance of OlmoEarth.

This project represents a powerful convergence of two major trends: the rise of foundation models and the urgent, global need for climate-change solutions.

For years, this kind of powerful, predictive environmental analysis was the exclusive domain of a few heavily-funded government agencies and multinational corporations. It was closed, proprietary, and slow. OlmoEarth shatters that model.

It’s a statement that the tools to save our planet shouldn’t be a trade secret.
It’s an investment in a future where a small conservation group in Madagascar, a university researcher in Brazil, and a municipal planner in California all have access to the same world-class AI to protect their communities.
It’s a bet that an open, collaborative community will always innovate faster, more creatively, and more equitably than any closed-door lab.

OlmoEarth has given us a new set of eyes. It has provided the “brain” to process what those eyes see. And it has made it all open and free for the world.

The only remaining question is: What will we choose to see? And what will we do about it?