Meta is putting a major new resource into the hands of materials researchers: Open Materials 2024, also called OMat24. The release includes a large data set and models designed to help scientists use AI to search for new materials faster.
The core issue is data. Modern materials discovery depends on simulations, calculations, and large training sets, but the source article describes those data sets as expensive to create, difficult to access, and often proprietary. Meta is making OMat24 free and open source, with the data set and models available on Hugging Face for anyone to download, test, and use.
Why materials discovery needs better data
Finding new materials is a computational problem as much as a scientific one. Researchers calculate the properties of elements across the periodic table, then simulate how different combinations may behave. That process can point toward materials with useful properties before researchers move further into development.
The potential uses are broad. The source article gives examples tied to climate change, including better batteries and new sustainable fuels. In those areas, the value of a new material can depend on discovering a combination of properties that is not obvious from existing options.
But the search is constrained by the scale and cost of the required data. Building large materials data sets takes substantial computing power. The article notes that many of the best data sets and models are proprietary, which means researchers cannot freely inspect, adapt, or build on them.
That is the bottleneck Meta says it is trying to address. Larry Zitnick, the lead researcher for the OMat project, describes the company’s view this way: “We’re really firm believers that by contributing to the community and building upon open-source data models, the whole community mov es further, faster,”
What Open Materials 2024 includes
Open Materials 2024 is both a data release and a model release. Meta says the OMat24 model will top the Matbench Discovery leaderboard, which ranks machine-learning models for materials science. The data set is also described as one of the biggest available.
According to the source article, Meta created the OMat24 data set by starting with an existing data set called Alexandria and sampling materials from it. The company then ran simulations and calculations involving different atoms to increase the scale.
The resulting data set has around 110 million data points. That makes it many times larger than earlier materials science data sets described in the source article. Shyue Ping Ong, a professor of nanoengineering at the University of California, San Diego, who was not involved in the project, says Meta has significantly expanded the data set beyond what the current materials science community has done, and with high accuracy.
The combination of size, open access, and claimed quality is why the release matters. A model can help researchers make predictions, but a large open data set can also become input for other models, comparisons, and experiments across the field.
How AI changes the materials workflow
The source article describes materials science as being in the middle of a machine-learning shift. Ong puts it plainly: “Materials science is having a machine-learning revolution,”
Before machine learning became more useful in this area, researchers faced a trade-off. They could run highly accurate calculations on very small systems, or less accurate calculations on very large systems. Both paths were described as laborious and expensive.
Machine learning helps bridge that gap. AI models can allow researchers to simulate combinations of any elements in the periodic table more quickly and cheaply, according to Ong. That does not remove the need for scientific judgment, but it changes how much ground researchers can cover computationally.
The article also points to a broader pattern: larger training sets can increase the potential to find new materials. Chris Bartel, an assistant professor of chemical engineering and materials science at the University of Minnesota, says tools such as Google’s GNoME (graphical networks for material exploration) have shown that this potential grows with training set size.
In practical terms, OMat24 is not only a single tool. It is a shared input that other scientists may use to evaluate ideas, build systems, and compare approaches in a field where access to data can shape who is able to participate.
Why open access is the central point
Several researchers quoted in the source article emphasize that the public release of the data set is more important than the model alone. Gábor Csányi, a professor of molecular modeling at the University of Cambridge, who was not involved in the work, says Meta’s decision contrasts with other large industry players.
He says: “This is in stark contrast to other large industry players such as Google and Microsoft, which also recently published competitive-looking models which were trained on equally large but secret data sets,”
That distinction matters because open data can be inspected, reused, and challenged. When a model is released without its underlying training data, outside researchers have less ability to understand what shaped its behavior or to adapt the resource for their own work.
The source article also notes that creating data sets of this size requires vast computational capacity. Meta is described as one of the few companies in the world that can afford that scale. The company also has its own reason to work on materials discovery: Zitnick says Meta hopes to find new materials that could make its smart augmented-reality glasses more affordable.
What researchers say could happen next
The history of open databases suggests that shared resources can reshape a research field. The source article cites previous open database work, including one created by the Materials Project, as having transformed computational materials science over the last decade.
Bartel frames OMat24 in that tradition. “The public release of the [OMat24] data set is truly a gift for the community and is certain to immediately accelerate research in this space,” he says.
The immediate significance is therefore not only that Meta has built a large AI materials data set. It is that the company is giving researchers access to a resource that would otherwise be difficult for many groups to create on their own.
If the promise described by the researchers holds, Open Materials 2024 could help more scientists participate in the search for useful new materials. The central bet is straightforward: when the data bottleneck gets smaller, the pace of discovery can get faster.