NVIDIA Introduces Nemotron for Science to Accelerate Discoveries With Agents

Glass Epoch DeskOriginal digestSource: NVIDIA Newsroom

AI-assisted digest, checked against the original source before publishing. How we work

NVIDIA packages Nemotron 3.5 Super VL with a 50,000-question chemistry dataset and post-training recipes so researchers can build specialized science agents.
NVIDIA Introduces Nemotron for Science to Accelerate Discoveries With Agents
Photo: Thufeil M · free stock license

NVIDIA announced Nemotron for Science, a bundle that pairs its open Nemotron 3.5 Super VL model with scientific evaluations, a new open chemistry dataset, and post-training recipes. The goal, per NVIDIA, is to help researchers develop agents that reason through scientific problems and put specialized research tools to work, adapting the model via post-training for tasks such as chemistry experiments and protein design.

The centerpiece dataset, LitmusBench v0.2, contains nearly 50,000 questions for training and evaluating language models on chemistry reasoning. It draws on molecular properties and experimental drug discovery data, and NVIDIA says it supports both evaluation and post-training, helping researchers gauge model capability and adapt models to scientific tasks. The dataset targets practical prediction problems: how molecules behave, including solubility and how the human body processes them.

Adoption is already underway across industry and national labs. Life sciences companies Lila Sciences, Terray Therapeutics and Therna are using and customizing Nemotron 3.5 Super VL for tasks including chemical knowledge extraction, experiment planning, and designing therapeutics and laboratory protocols. On the research side, the U.S. Department of Energy's Argonne National Laboratory and Jülich Supercomputing Centre are integrating the model into drug and scientific discovery AI agents.

NVIDIA framed the release around infrastructure reach as much as model capability. DGX Spark and data center systems are positioned to carry Nemotron 3.5 Super VL from researchers' desktops to supercomputing centers, enabling secure scientific workflows on infrastructure organizations control. Ian Buck, NVIDIA's VP of hyperscale and high-performance computing, said the package gives researchers "open models, scientific data and post-training recipes to build specialized agents that reason through scientific problems and use CUDA-X accelerated tools to turn ideas into discoveries."

Post-trained models are meant to slot into agent workflows that interpret scientific information, generate code, and call computational tools, with NVIDIA's CUDA-X stack handling the underlying scientific computation. That combination positions the release at the intersection of open model weights, domain datasets, and agent orchestration rather than a single benchmark-topping model.

Open questions remain around the practical impact. It is not yet clear how much post-training on LitmusBench v0.2 shifts real laboratory outcomes versus benchmark scores, and how institutions without DGX-class infrastructure will run the same secure workflows. Watch for follow-up results from Argonne, Jülich, and the three life sciences companies on whether these agents shorten experiment cycles in production settings.

Key facts

  • • LitmusBench v0.2 ships with nearly 50,000 chemistry reasoning questions.
  • • Nemotron for Science builds on the open Nemotron 3.5 Super VL model.
  • • Argonne National Laboratory and Jülich Supercomputing Centre are integrating the model into discovery agents.
  • • Lila Sciences, Terray Therapeutics and Therna are customizing the model for science applications.
  • • NVIDIA positions DGX Spark to move the model from desktops to supercomputing centers.

Why it matters

The real asset here is the pairing of an open model with a domain-specific dataset and training recipes, which lowers the bar for labs to specialize agents rather than starting from scratch. Builders should watch whether post-training on LitmusBench translates into shorter experiment cycles at Argonne and the adopting life sciences firms — and whether institutions lacking DGX-class hardware can re

Spotted an error? Send a correction.

Get the day's AI news in one email

The Daily Epoch · 60-second read · unsubscribe anytime

Related