News

NVIDIA Launches Nemotron for Science to Advance Research with AI Agents

NVIDIA introduces Nemotron for Science, combining AI models, datasets, and tools to accelerate discoveries in life sciences and chemistry.

NVIDIA Launches Nemotron for Science to Advance Research with AI Agents

NVIDIA has unveiled Nemotron for Science, a comprehensive platform designed to accelerate scientific discovery by integrating its Nemotron 3.5 Super VL model with new tools and datasets. The initiative aims to help researchers develop domain-specific AI agents for applications in life sciences, chemistry, and materials science.

Key Features of Nemotron for Science

The platform combines the Nemotron 3.5 Super VL open model with a new chemistry-focused dataset, LitmusBench v0.2, and post-training recipes. LitmusBench v0.2 includes nearly 50,000 questions for evaluating and training language models on chemistry reasoning. By leveraging molecular properties and experimental drug discovery data, the dataset enhances models' ability to predict molecular behavior, such as solubility and human body processing.

Researchers can post-train Nemotron models with domain-specific data to improve performance on specialized tasks. These tasks include conducting chemistry experiments, designing proteins, and creating synthesis recipes for new materials. NVIDIA’s CUDA-X accelerated tools further enable scientific computation within agent workflows.

Applications and Early Adopters

Several organizations are already utilizing Nemotron for Science. The U.S. Department of Energy’s Argonne National Laboratory is integrating the platform into its StructBioReasoner multi-agent system to design protein-based therapeutic candidates for cancer treatment. KTH Royal Institute of Technology has developed an AI agent using Nemotron to autonomously repair GROMACS molecular dynamics setups, achieving an 81% success rate in overnight tests.

Life sciences companies such as Lila Sciences, Terray Therapeutics, and Therna are also customizing Nemotron for various applications. For instance, Terray Therapeutics has used the platform to extract reaction conditions from scientific diagrams, achieving a 76% recovery rate on the RxnScribe benchmark. Lila Sciences has improved quantum dot synthesis accuracy to 61%, surpassing GPT-6 Astra’s 54% at significantly lower inference costs. Therna has nearly doubled the accuracy of mRNA therapeutic design for protein-expression prediction, with scientists favoring its explanations in 96% of comparisons.

Infrastructure and Accessibility

NVIDIA’s DGX Spark systems and data center infrastructure enable Nemotron for Science to scale from individual researchers’ desktops to supercomputing centers. RIKEN has made Nemotron available on its RIKYU AI supercomputing platform, providing researchers with access to advanced models and tools for scientific research.

Nemotron 3.5 Super VL and the LitmusBench v0.2 dataset are now available on Hugging Face, with additional technical resources provided by NVIDIA to support post-training and deployment.

Sign in to save this article Free account: build your research collection and return to read it anytime.