NVIDIA Nemotron for Science: Open Models Pull Labs Onto NVIDIA Silicon
As seen on the 24/7 Wall St. homepage on November 16, 2026.
NVIDIA Introduces Nemotron for Science to Accelerate Discoveries With Agents
Open models are becoming the hook that pulls national labs and biotech onto Nvidia silicon: Argonne, Jülich, RIKEN, Lila Sciences, Terray and Therna are all building agents on Nemotron 3.5 Super VL. Lila's post-trained model hit 61% accuracy designing quantum dot recipes, ahead of GPT-6 Astra at far lower inference cost. The free dataset sells the hardware underneath it.
Continue ReadingShow less
NVIDIA Nemotron for Science bundles the Nemotron 3.5 Super VL open model with a new open chemistry dataset and post-training recipes, giving research institutions a ready-made path to building specialized AI agents for scientific discovery. The package is free to use, which matters because the dataset is effectively a loss leader designed to route compute demand toward NVIDIA hardware.
The early adopter list reads like a who's who of serious scientific compute: national laboratories at Argonne, Jülich, and RIKEN, alongside biotechs Lila Sciences, Terray, and Therna, are all building agents on the platform. That breadth of institutional commitment across both publicly funded labs and private biotech suggests the strategy is gaining traction well beyond a proof-of-concept stage.
Sponsored
The clearest benchmark result in the launch comes from Lila Sciences, whose post-trained model built on Nemotron 3.5 Super VL hit 61% accuracy designing quantum dot recipes. That figure beat GPT-6 Astra on the same task and did so at lower inference cost, which is the combination that tends to shift procurement decisions at organizations running experiments at scale.
For NVIDIA as a hardware business, the logic is straightforward: open, freely available models lower the barrier to entry for scientific institutions, expand the installed base of researchers building on NVIDIA silicon, and create a pipeline of inference workloads that eventually need more GPUs. The chemistry dataset and post-training recipes draw institutions in, and the accelerators are what they ultimately buy.
Mentioned: NVDA