NVIDIA Launches Nemotron for Science to Accelerate Scientific Discoveries with Intelligent Agents
NVIDIA today launched NVIDIA Nemotron for Science, which integrates the scientific applications of the NVIDIA Nemotron™ 3.5 Super VL open model, a new open chemistry dataset, and post-training recipes to help researchers build dedicated intelligent agents for scientific discovery.
Built on Nemotron 3.5 Super VL, Nemotron for Science assists researchers in developing intelligent agents capable of reasoning about scientific problems and invoking specialized research tools. Through post-training, scientists can adapt the model for specialized tasks such as chemical experiments and protein design.
As part of Nemotron for Science, LitmusBench v0.2 is a brand-new open dataset containing nearly 50,000 questions for training and evaluating the chemical reasoning capabilities of language models. This dataset is based on molecular properties and experimental drug discovery data, helping researchers enhance the model's ability to predict molecular behaviors, including solubility and how the human body metabolizes these molecules.
Life sciences companies Lila Sciences, Terray Therapeutics, and Therna are using and customizing NVIDIA Nemotron 3.5 Super VL for scientific applications. Use cases include extracting chemical knowledge, planning experiments, and designing therapies and laboratory protocols. Major research centers such as Argonne National Laboratory and Jülich Supercomputing Centre are integrating it into their drug and scientific discovery AI agents.
"Accelerated computing is changing the way scientists tackle the world's most complex challenges," said Ian Buck, NVIDIA's Vice President of Hyperscale and High-Performance Computing. "Nemotron for Science provides researchers with open models, scientific data, and post-training recipes, enabling them to build dedicated intelligent agents that reason about scientific problems and use CUDA-X acceleration tools to turn ideas into discoveries."
Post-Training of the Nemotron Open Model for Scientific Discovery
Researchers can use domain-specific data to post-train the Nemotron open model to improve performance on specialized scientific tasks. They can integrate the resulting model into intelligent agent workflows for interpreting scientific information, generating code, and invoking computational tools. These workflows can execute scientific computations using NVIDIA's CUDA-X™ acceleration tools.
Terray Therapeutics has post-trained Nemotron to extract reactants, products, and reaction conditions from scientific charts, surpassing previously published results on two benchmark tests. In the RxnScribe test, it recovered about 76% of the annotated reactions according to the benchmark's matching criteria, aiding in converting published chemical discoveries into structured data for synthesis planning and scientific agents.
Lila Sciences uses Nemotron to design executable laboratory protocols for its automated laboratory, achieving cutting-edge levels without additional fine-tuning. After post-training with laboratory feedback, the model's accuracy in producing quantum dots (nanoscale crystals for screens and solar panels) nearly doubled to 61%, surpassing GPT-6 Astra's 54%, with reasoning costs reduced by 10 to 20 times. These improvements help intelligent agents and materials scientists translate desired material properties into synthesis recipes, driving the discovery of new materials.
Therna collaborated with NVIDIA to post-train Nemotron for intelligent agent-based mRNA therapy design, nearly doubling accuracy in predicting protein expression ranges, surpassing state-of-the-art models. In 96% of comparisons, scientists preferred its explanations over those of the baseline model. These enhancements can help scientists customize mRNA designs for specific cell types to achieve desired protein yields, extending to human applications.
Researchers Accelerate Scientific Research with Nemotron Models and Datasets
Leading research institutions are using NVIDIA Nemotron models and datasets to advance scientific discoveries at various scales, from individual researchers' desktops to laboratories and supercomputing centers.
By helping researchers quickly design and evaluate new molecules from the periodic table, this work aims to open new pathways for critical materials and minerals, national security, and cancer treatment applications.
Argonne National Laboratory is integrating Nemotron into its StructBioReasoner multi-agent system for designing and evaluating protein therapy candidates for cancer and other diseases.
The Royal Institute of Technology in Sweden has built an AI agent based on Nemotron that can autonomously repair damaged GROMACS molecular dynamics setups. In a single overnight run on an NVIDIA DGX Spark™ system, the agent repaired 81% of test cases while reasoning, with each repair independently verified by GROMACS itself.
The RIKEN Center for Computational Science in Japan has deployed NVIDIA Nemotron, including Nemotron 3.5 Super VL, on its RIKYU AI supercomputing platform through its reasoning gateway RiVault, enabling researchers to access these models for scientific research and accelerate application development.
Availability
NVIDIA Nemotron 3.5 Super VL and the LitmusBench v0.2 chemistry dataset are now available on Hugging Face. The Nemotron 3.5 Super VL model has also been optimized for local execution on a 128GB NVIDIA DGX Spark system.
-- Price
This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.
You may also like

AI Fills Gaps in the Crypto Industry: Fifteen Startup Opportunities to Watch

Blockchain.com Adds Former Microsoft Executive and Global X ETFs CEO to Board

Spark's Q3 Lending Market Share Rises to 13.2%, Net Income Drops 30%

The Evolution of Ethereum Staking: Treehouse tETH, a New Infrastructure in the DeFi Interest Rate Market
![[Column] Robinhood Chain: The Meme Coin Craze May Be Over, But Growth Continues](/public-static/075_ade8940e45.png?format=avif)
[Column] Robinhood Chain: The Meme Coin Craze May Be Over, But Growth Continues

Delphi Digital: Why Speculation is the Theme of the Next Decade

Julien Vallet (Finst) on Legend: He Reveals His Wildest Crypto Stories

Programmable Trading Arrives: How Hook Becomes the 'Brain' of the Trading Pool?

BlackRock CEO Larry Fink Flags a Different AI Risk, and It Isn’t the Bubble

Meta Releases 'America's Strongest Open Source Model'

Meta Releases Open Weight AI Model Muse Spark Open

ICO, VC Tokens, and Alpha Models: Why They Can't Save the Current Crypto Market?

Stricter Regulation May Lead Brazilian Crypto Market to a New Concentration Cycle

Spark Reports Over $4 Million in Net Protocol Returns for Q3, First Month Transaction Volume Exceeds $14 Billion

Meta Launches Muse Wonder Series Models, Claims Leading Performance in Agent Tasks

Crypto VC funding: Spiko and Nous Research raise $90m each

Breez Reports 14-Fold Increase in Bitcoin Integration Demand

US Stocks Can Be Pledged: The On-Chain Lending Loop Still Needs a Few More Links

Wall Street Morning Report: AI Bull Market Faces Interest Rate Judgment, Funds Shift to Defensive Assets, 30-Year Treasury Bonds Become Next Test

Lightspark Launches dotMoney Prototype, Supporting Link and .money File Transfers for Test Funds

Google Launches Playground to Generate Playable Games

Spark Leads $2.4 Million Investment as Antseed Launches AI Inference Network

Who Will Aave Hand Its Brand Over To? The DAO Asset Ownership Dispute Behind the Foundation Proposal

Polymarket: Bets on Bank Failures Spark Reactions in London

Ripple RLUSD Supply Increased by $765 Million in Q3

RWA Weekly: Hong Kong SAR Government Introduces HKD Tokenized Deposits; Stripe-led USD Stablecoin OUSD Officially Launched

Why Did Google Stock Reverse After Gemini 4 Argon: What's Actually Going On

Is the Compound Foundation 'Self-Theft'?

Statement on the Misreporting of the Fomo App by BlockBeats










