Nemotron for Science: NVIDIA's open model, chemistry dataset and recipes for science agents
NVIDIA packaged its open Nemotron 3.5 Super VL vision-language model with the LitmusBench v0.2 chemistry dataset (about 50,000 questions) and post-training recipes so researchers can build domain-specific science agents.
Why it matters
The first open bundle that pairs a 120B multimodal model with a real chemistry training set and the recipes to adapt it, plus named deployments at national labs and biotechs showing it working on actual science, not demos.
What you could build with it
A university chemistry lab could post-train the open model on its own assay data using the LitmusBench recipes, then run a small agent that reads new papers, proposes candidate molecules, and checks each one against the lab's own experimental records before a human reviews the shortlist.
Does it hold up?
Too early to judge as a bundle: the model and dataset are live on Hugging Face, but the named deployments are vendor-reported case studies and no independent evaluation of the science claims exists yet.
Built with Nemotron for Science
- KTH's GROMACS repair agent fixed 81% of broken simulation setups overnight on one DGX Sparkorcarouter · KTH Royal Institute of Technology built an agent on Nemotron for Science that repaired deliberately broken molecular-dynamics setups, with every fix verified by GROMACS itself.
- Argonne integrates Nemotron into StructBioReasoner for protein therapeutic designtechstrong · Argonne National Laboratory is building StructBioReasoner, a multi-agent system that designs and evaluates protein-based therapeutic candidates, on Nemotron 3.5 Super VL.
- Terray Therapeutics post-trains the model on chemistry diagramsglobenewswire · Terray Therapeutics is customizing Nemotron 3.5 Super VL on chemistry diagrams as part of the launch's named life-sciences deployments.
- nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16github · The open 120B-parameter multimodal model weights on Hugging Face: 12B active per token, up to 1M context, text, images and short video in.
- nvidia/Nemotron-RL-litmus-bench-v0.2github · LitmusBench v0.2: about 50,000 short-answer chemistry questions for training and evaluating models on chemistry reasoning.
Learn more
First spotted on article: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73Google AI Edge Gallery: offline speech recognition and synthesis on-deviceGoogle added fully offline speech recognition and text-to-speech to AI Edge Gallery, with ML Kit APIs on Android and…models · JEV 0.73Soup: fine-tune an 8B LLM on a 4 GB laptop GPU from one YAML fileAn open-source CLI that fine-tunes LLMs from a single YAML file, with layer streaming that trains an 8B model on a 4…models · JEV 0.73Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.