Nvidia launches Nemotron for Science at SC26 with open chemistry data for research agents

Nvidia on 16 November unveiled Nemotron for Science at the SC26 supercomputing conference in Chicago: built around the open Nemotron 3.5 Super VL model, it adds the LitmusBench v0.2 dataset of nearly 50,000 chemistry reasoning questions and post-training recipes to help researchers build agents for scientific discovery.
Nvidia on 16 November unveiled Nemotron for Science at the SC26 supercomputing conference in Chicago. Built around the open Nemotron 3.5 Super VL model, the package adds the LitmusBench v0.2 dataset of nearly 50,000 chemistry reasoning questions and post-training recipes, to help researchers build agents for scientific discovery across chemistry, life sciences, protein design and materials science.
What was released
Nemotron 3.5 Super VL is an open-weight multimodal model Nvidia released in October: a mixture-of-experts design with 120 billion total parameters and 12 billion active per token, able to work with text, images, documents and video. LitmusBench v0.2 is a new open dataset whose nearly 50,000 questions draw on molecular properties and experimental drug-discovery data, for training and testing language models on chemistry reasoning, such as predicting how molecules dissolve and how the body processes drugs. Nvidia also supplies post-training recipes so research teams can adapt the model with their own data, tools and workflows. The model and dataset are available now on Hugging Face, and the model is optimised to run locally on a 128GB DGX Spark system.
Early users
Nvidia named life sciences companies Lila Sciences, Terray Therapeutics and Therna among its early users, alongside the US Department of Energy's Argonne National Laboratory and Germany's Juelich Supercomputing Centre. The company said the package builds on its earlier life sciences project BioNeMo and plugs into the CUDA-X accelerated computing suite, so it can be deployed across different computing environments.
Rubin HPC announced the same day
Nvidia also used SC26 to announce Rubin HPC, a new graphics processor in its Vera Rubin platform aimed at scientific simulation and data processing, expected to be available in 2028 with no price given — a long-term roadmap announcement rather than a product going on sale soon. The figures and user list above are Nvidia's own claims, and independent evaluation is still to come.
N環球 
💬 Comments
No comments yet. Be the first!