MatterChat AI helps scientists find new materials on 19 May 2026

The new MatterChat AI model can now see atoms to help build better materials. This is a big step forward compared to old text-only AI models.
By Newsroom | Technology, Science |

A new AI model, dubbed MatterChat, is being presented as a bridge, enabling text-based artificial intelligence to interpret the complex forces between atoms. This development, originating from Lawrence Berkeley National Laboratory (Berkeley Lab), aims to improve the prediction of new materials by granting AI a form of structural "vision."

The MatterChat model draws inspiration from vision question answering and text-to-image technologies. Its core function is to translate the intricacies of atom-scale physics into a format that AI can process, moving beyond mere data analysis to a deeper, structural understanding. This approach is designed to be "forward-compatible," meaning it can integrate future, larger datasets and advancements in AI.

MatterChat model helps AI to 'see' the language of atom-scale physics to sharpen materials predictions - 1

The project leveraged supercomputing resources from the National Energy Research Scientific Computing Center (NERSC), a facility managed by Berkeley Lab for the U.S. Department of Energy. Key contributors include Wenbin Xu, a former NERSC postdoctoral fellow, and Benjamin Erichson, a research scientist at Berkeley Lab's Scientific Discovery Software Division (SDD), underscoring a collaborative effort within the lab.

Read More: New Ge-Si Photodetector Reaches 336 Gbps for AI Data Centers

MatterChat is described as a multimodal large language model (LLM). It has undergone extensive training not only on scientific literature but also on scientific imagery and crystal structure databases. This training allows it to perform tasks such as analyzing crystal structures and predicting material properties. The model is built upon the Qwen-VL architecture, adapted specifically for the domain of materials science.

MatterChat model helps AI to 'see' the language of atom-scale physics to sharpen materials predictions - 2

The development represents a shift in how AI is applied to scientific challenges, focusing on creating specialized "connective tissue" rather than solely building larger, general-purpose AI models. This specialized approach seeks to make commercial AI more useful for rigorous scientific inquiry.

Read More: Mars solar flare on 15 May 2026 triples ionosphere electron density

Background and Context

Berkeley Lab is a significant player in several U.S. Department of Energy (DOE) Energy Innovation Hubs and hosts various research institutes. NERSC, as the DOE Office of Science's mission computing facility, supports a broad spectrum of scientific research, including materials science, physics, and chemistry. The ongoing discourse in the field includes research into foundational LLMs for materials, understanding hallucinations in multimodal LLMs, and applying LLMs to synthesis predictions.

Frequently Asked Questions

Q: What is the new MatterChat AI tool released by Berkeley Lab?
MatterChat is a new AI model that can 'see' atoms and understand how they fit together. It helps scientists predict the properties of new materials by analyzing complex crystal structures.
Q: How does MatterChat help materials scientists today?
It allows researchers to use text-based AI to interpret atom-scale physics. This makes it easier and faster for scientists to find new materials for energy and technology.
Q: Who created the MatterChat AI model?
The model was created by researchers Wenbin Xu and Benjamin Erichson at the Lawrence Berkeley National Laboratory. They used supercomputers from NERSC to train the AI on scientific images and data.
Q: Why is MatterChat different from other AI models?
Most AI models only read text, but MatterChat is a multimodal model that understands scientific images and crystal structures. It is built to grow and handle even larger amounts of data in the future.