Skip to main content
Live
Main content

Nvidia launches Nemotron for Science with open chemistry dataset for research agents

Early users report 76% reaction recovery and a 61% quantum-dot synthesis rate, and the model runs locally on a 128GB DGX Spark.

Jaeden Schafer
Editor in Chief · · 4 min read
Nvidia logo

Nvidia introduced Nemotron for Science on November 16, a package that pairs its open Nemotron 3.5 Super VL model with a new open chemistry dataset and post-training recipes for building scientific-discovery agents. Nvidia said early users have applied it to extracting reactions from chemistry diagrams, writing protocols for automated labs and designing mRNA therapeutics, with one partner reporting 61% accuracy on quantum-dot synthesis.

The centerpiece dataset is LitmusBench v0.2, which Nvidia says holds nearly 50,000 questions for training and evaluating language models on chemistry reasoning. It draws on molecular property and experimental drug-discovery data, covering tasks such as predicting solubility and how the human body processes a molecule. Both the dataset and Nemotron 3.5 Super VL are available now on Hugging Face.

Nvidia's pitch is that researchers can post-train Nemotron open models on their own domain data, then wire them into agent workflows that interpret scientific information, generate code and call tools, including Nvidia's CUDA-X accelerated libraries for scientific computation. The company named extracting chemical knowledge, planning experiments, and designing therapeutics and laboratory protocols as target uses.

Key facts

  • 01Nvidia introduced Nemotron for Science on Nov. 16, built on its open Nemotron 3.5 Super VL model.
  • 02LitmusBench v0.2, a new open chemistry dataset, contains nearly 50,000 questions; it and the model are on Hugging Face.
  • 03Lila Sciences reached 61% quantum-dot synthesis accuracy vs. 54% for GPT-6 Astra, at 10-20x lower inference cost, per Nvidia.
  • 04Terray Therapeutics recovered about 76% of annotated reactions on the RxnScribe benchmark.
  • 05The model is optimized to run locally on a 128GB DGX Spark; KTH's agent fixed 81% of GROMACS test cases overnight on one.

The model was also optimized to run locally on a 128GB DGX Spark system. Nvidia positions DGX Spark alongside its data center infrastructure so the same Nemotron 3.5 Super VL model can move from a researcher's desktop to a supercomputing center. Ian Buck, Nvidia's vice president of hyperscale and high-performance computing, framed the release around agents that reason through scientific problems and use accelerated tools to turn ideas into discoveries.

“Accelerated computing is transforming how scientists tackle the world's most complex challenges”
— Ian Buck, Vice president of hyperscale and high-performance computing at Nvidia

Terray Therapeutics post-trained Nemotron to pull reactants, products and reaction conditions out of scientific diagrams. Nvidia says the result surpassed previously published results on two benchmarks, recovering about 76% of annotated reactions on the RxnScribe test under that benchmark's matching criteria.

Lila Sciences used Nemotron to design execution-ready protocols for automated labs. According to Nvidia, the model matched frontier-level performance without additional fine-tuning, and post-training with lab feedback nearly doubled accuracy on quantum-dot synthesis to 61%. Nvidia compares that with 54% for GPT-6 Astra, at 10 to 20 times lower inference cost.

Therna worked with Nvidia to post-train Nemotron for agentic mRNA therapeutic design, nearly doubling accuracy in predicting protein-expression ranges compared with frontier models, the company said. Scientists preferred the tuned model's explanations over the base model's in 96% of comparisons.

At KTH Royal Institute of Technology, an agent built on Nemotron repairs broken GROMACS molecular dynamics setups. In one overnight run on a single DGX Spark, it fixed 81% of test cases with reasoning enabled, and GROMACS independently checked every fix.

Related · from this week
Nvidia ships Nemotron 3.5 Lightning, a 30B open MoE for local agents
Jaeden Schafer · 5 min read →

The U.S. Department of Energy's Argonne National Laboratory is integrating Nemotron into its StructBioReasoner multi-agent system to design and evaluate protein-based therapeutic candidates for cancer and other diseases. RIKEN has made Nemotron models, including Nemotron 3.5 Super VL, available on its RIKYU AI supercomputing platform through its RiVault inference gateway, and the Jülich Supercomputing Centre is integrating Nemotron into drug and scientific discovery agents.

The performance figures in the announcement are reported by Nvidia and its partners, and the release does not include independent verification. Nvidia published a companion technical blog post on post-training Nemotron to raise accuracy in specific scientific domains.

ShareXLinkedInEmail
AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

  • ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat
  • Generate images & video with Sora, Veo, Ideogram
  • Compare any two models side by side
  • From $8.99/mo · 80+ models, all included
Try AI Boxaibox.ai
Trusted by 3,000+ teams
Got a tip?

Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.

Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.

AI Box Daily briefingFree · Daily · No fluff

Stay ahead of everyone in AI.

The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.

Loved by 10,000+ AI professionals
Free forever. Unsubscribe with one click.

The briefing read inside teams at

Keep reading

More from Models

Nvidia logo
Models

Nvidia ships Nemotron 3.5 Lightning, a 30B open MoE for local agents

The new open-weights model runs 4x faster than class rivals and slots into RTX PCs, DGX Spark, and Jetson for always-on agentic workloads.

Jaeden Schafer5 min read
Nvidia logo
Models

Nvidia launches $4,999 DGX Spark 64GB for local AI development

The new SKU runs 100B-parameter models on device; two units cluster to 128GB and 200B via Nvidia Sync Cluster Assistant.

Jaeden Schafer4 min read
Nvidia logo
Models

Manchester researchers use Nvidia Earth-2 to forecast UK air pollution in two days

A generative downscaling model trained on Isambard-AI now runs UK-wide pollution forecasts at 2-3 km resolution, with inference on a desktop.

Jaeden Schafer5 min read