Background
General-purpose language models are trained on broad text, but scientific work often needs knowledge of specialized fields such as chemistry or biology, along with ways to test whether a model is accurate in those fields. Researchers who want to build such models need open data, evaluation tools and training methods they can reuse.
What happened
On November 16, 2026, NVIDIA introduced Nemotron for Science to accelerate discovery with AI agents.[1] The package combines NVIDIA's Nemotron 3.5 Super VL open model, scientific evaluations, a new open chemistry dataset and post-training recipes.[1] Post-training means the additional training a model receives after its initial pre-training, usually to make it better at specific tasks.
Key details
The chemistry dataset is LitmusBench v0.2, an open dataset of nearly 50,000 questions. It is designed for training and evaluating language models on chemistry reasoning, including solubility and how the body processes molecules.[1]
Several organizations are already using or customizing Nemotron 3.5 Super VL for science. Lila Sciences, Terray Therapeutics and Therna are among them.[1]
Argonne National Laboratory is integrating Nemotron into its StructBioReasoner multi-agent system, which is used for designing protein-based therapeutic candidates.[1] A multi-agent system is a set of AI agents that divide a task and pass results to one another.
The Jülich Supercomputing Centre is integrating the model into its drug and scientific discovery AI agents.[1]
Open models and open datasets let other groups inspect, rerun and adapt the same work. A question set such as LitmusBench is useful to the wider field only if other groups can run the same questions and compare their results with the published ones.
Impact
The package gives research groups an open model, a test set and training recipes in one place. The examples in the announcement cover drug design, protein work and chemistry, which are fields where the evaluation data and the model are meant to be used together.
Sources
- 1
Cite this page: NVIDIA introduces Nemotron for Science. TellWorld, https://www.tellworld.net/news/nvidia-nemotron-for-science-2026. Updated 2026-11-18