Nearly 145 papers accepted at this year's International Conference on Machine Learning cite NVIDIA's Nemotron open models and datasets as a foundation for new research, according to an NVIDIA blog post published on 6 July. NVIDIA itself had 74 papers accepted, while roughly 2,000 of the conference's accepted papers reference NVIDIA GPUs.
The post frames Nemotron less as a single release and more as a research stack, offering open weights, datasets and recipes for reasoning, tool use, safety, data curation and efficient inference. Beyond Nemotron, hundreds of papers drew on NVIDIA's Cosmos, Isaac GR00T and BioNeMo model families, spanning robotics, autonomous vehicles and biomedical research.
Adoption beyond NVIDIA's own labs is notable. Sakana AI built its Fugu and Fugu-Ultra models on Nemotron 3 Ultra, and NAVER developed its own model on the Nemotron architecture for Korean-language research. KiloCode, meanwhile, claims that integrating Nemotron into its code-routing architecture cut token costs by up to 90 per cent, though this figure is self-reported and has not been independently verified. Merck & Co. is using NVIDIA's KERMT model to predict how drug candidates may behave in the body.
The results suggest open infrastructure is becoming the default substrate for a growing share of frontier AI research, rather than a niche alternative to closed models.
