NVIDIA packages Nemotron models and chemistry data for scientific agents

NVIDIA packages Nemotron models and chemistry data for scientific agents
By guthlabs.newsroom · November 16, 2026
NVIDIA introduced Nemotron for Science at SC26 on November 16, combining its Nemotron 3.5 Super VL model with scientific evaluations, chemistry data and post-training recipes. The company says the model and LitmusBench v0.2 dataset are available on Hugging Face. Official announcement
LitmusBench contains nearly 50,000 questions for training and assessing chemistry reasoning. NVIDIA describes applications including extracting reaction information, planning experiments and adapting agents to specialized scientific tasks. The release also says the model has been optimized for a 128GB DGX Spark, offering a local deployment option alongside larger research infrastructure.
The announcement reports partner results, including an agent that repaired 81% of broken GROMACS test setups during an overnight run. These are reported experimental outcomes, not independently reproduced guarantees of performance on other laboratories’ systems.
Guth News analysis: The useful distinction is between a general model and a domain workflow with relevant training data, tools and checks. This package makes that adaptation more concrete. It does not remove the need for held-out evaluation, reproducible experiments or human responsibility for laboratory actions. A successful software repair or benchmark answer is not itself a validated scientific discovery.
Evidence note: Event and source date November 16. Primary announcement full text read. Model and dataset availability attributed to NVIDIA; direct checkpoint and license verification not completed, and no license claim made. No independent reproduction located.
Sources and citations
Revision history
Published .