ICML 2026: open models become the foundation of AI research

The International Conference on Machine Learning (ICML) 2026 took place in Seoul, South Korea, with open models and open AI infrastructure emerging as dominant themes.

NVIDIA reported that 74 of its own papers were accepted at the conference. Approximately 2,000 accepted papers cite NVIDIA GPUs as infrastructure, and 145 papers cite NVIDIA Nemotron models and datasets as a foundation for new research.

Additional papers drew on NVIDIA Cosmos, Isaac GR00T, BioNeMo and other open model families, spanning physical AI, robotics, autonomous vehicles, and biomedical research.

The numbers underscore a broader trend: open-weight models and open datasets are becoming the default building blocks for modern AI research, enabling reproducibility and lowering the barrier to entry for labs without the resources to train frontier models from scratch.