
The package pairs an open model with a chemistry dataset and training recipes, while research groups report early applications.
NVIDIA introduced Nemotron for Science, a package for building specialized AI agents for research, on Nov. 16. It combines the Nemotron 3.5 Super VL model with training recipes and LitmusBench v0.2, an open chemistry dataset with nearly 50,000 questions.
The dataset is designed to train and evaluate models on chemistry reasoning, including predictions about molecular behavior. NVIDIA said the model and dataset are available on Hugging Face.
Research groups are applying the model to different tasks. Argonne National Laboratory is integrating it into an agent system for designing and evaluating protein-based therapeutic candidates. KTH Royal Institute of Technology built an agent that repaired 81% of test cases for broken molecular dynamics setups during an overnight run, NVIDIA said.
Companies also reported results from customizing the model. Terray Therapeutics said it recovered about 76% of annotated reactions on the RxnScribe test. Lila Sciences reported that post-training with laboratory feedback raised the model’s accuracy in making quantum dots to 61%.
NVIDIA said the system can be adapted for tasks such as chemistry experiments and protein design, with workflows that use computational tools. Research organizations including Argonne and Jülich Supercomputing Centre are integrating Nemotron into scientific discovery agents.
This article was produced with the help of AI technology. Based on reporting by the outlets listed above.