NVIDIA launches Nemotron for Science for AI-driven research workflows

NVIDIA launches Nemotron for Science for AI-driven research workflows
NVIDIA powers AI science

AI tools for drug discovery, chemistry and materials research are drawing increased investment as laboratories seek faster ways to turn scientific data into usable results. NVIDIA is introducing Nemotron for Science with an open model, a chemistry dataset and post-training recipes aimed at helping researchers build specialized agents for scientific discovery.

Highlights

  • NVIDIA launched Nemotron for Science, integrating Nemotron 3.5 Super VL, LitmusBench v0.2, and domain-specific recipes to accelerate AI-powered chemistry and life science research workflows.
  • Terray Therapeutics post-trained Nemotron, achieving 76% reaction recovery on RxnScribe and surpassing past benchmarks in extracting structured data from scientific chemistry diagrams.
  • NVIDIA reported that post-training with laboratory feedback doubled Nemotron model accuracy for quantum dot creation to 61% versus GPT-6 Astra’s 54%, while inference cost dropped 10-20 times.

Launch package targets chemistry and life sciences

As announced by NVIDIA Newsroom, the new NVIDIA Nemotron for Science brings together scientific applications of the company’s Nemotron 3.5 Super VL open model, the LitmusBench v0.2 open chemistry dataset and post-training recipes for domain-specific research tasks.

Built on Nemotron 3.5 Super VL, the offering is designed to help researchers create agents that reason through scientific problems and use specialized tools in workflows such as chemistry experiments, protein design and code generation. LitmusBench v0.2 includes nearly 50,000 questions for training and evaluating language models on chemistry reasoning, drawing on molecular properties and experimental drug discovery data to improve predictions on molecule behavior, including solubility and how the human body processes them.

NVIDIA says life sciences companies including Lila Sciences, Terray Therapeutics and Therna are using and customizing the model for tasks such as extracting chemical knowledge, planning experiments, designing therapeutics and generating laboratory protocols. Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA, says the package gives researchers open models, scientific data and post-training recipes that can be combined with CUDA-X accelerated tools to turn ideas into discoveries.

Early users report gains in model accuracy and lab automation

Terray Therapeutics post-trained Nemotron to extract reactants, products and reaction conditions from scientific diagrams, surpassing previously published results on two benchmarks. On the RxnScribe test, the model recovered about 76% of annotated reactions under the benchmark’s matching criteria, which can help convert published chemistry findings into structured data for synthesis planning and scientific agents.

Lila Sciences used Nemotron to design execution-ready laboratory protocols for automated labs, matching frontier-level performance without additional fine-tuning. NVIDIA says post-training with laboratory feedback nearly doubled the model’s accuracy for making quantum dots to 61%, above GPT-6 Astra’s 54%, while operating at 10-20 times lower inference cost.

Therna collaborated with NVIDIA to post-train Nemotron for agentic mRNA therapeutic design, nearly doubling accuracy in predicting protein-expression ranges over frontier models. Scientists preferred its explanations over those from the base model in 96% of comparisons, a result that could support more tailored mRNA designs for specific cell types.

Research institutions are also integrating the models into scientific computing platforms and agent systems. Argonne is adding Nemotron to its StructBioReasoner multi-agent system for protein-based therapeutic design, KTH Royal Institute of Technology has built an agent that repairs broken GROMACS molecular dynamics setups, and RIKEN is making Nemotron models available through its RIKYU AI supercomputing platform via the RiVault inference gateway.

We recently reported on YTL Power scrutiny tied to Nvidia-related pressure and its Johor AI data center project in Malaysia.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.