NVIDIA Dominates ICML 2026: 74 Papers Accepted, 145 Cite Open Nemotron Models

NVIDIA had 74 papers accepted at the International Conference on Machine Learning (ICML) 2026, according to a blog post published Monday by NVIDIA content specialist JJ Kim. Approximately 2,000 accepted papers cite NVIDIA GPUs, and 145 papers cite NVIDIA Nemotron — a family of open models including open datasets — as the foundation for new research. Hundreds more draw on NVIDIA Cosmos, Isaac GR00T, BioNeMo, and other NVIDIA open model families.

The conference results reveal that open frontier models and open AI infrastructure have become foundational to how modern AI research gets done. The open model ecosystem spans physical AI, robotics, autonomous vehicles, and biomedical research.

Key research themes at ICML 2026 include vision and video generation, reinforcement learning for large language models (LLMs), and agent training. Robot world models emerged as a breakout area, exemplified by DreamDojo, which learns physical world behavior from human video and builds on NVIDIA Cosmos to predict how robots would handle objects in unseen environments.

In life sciences, NVIDIA BioNeMo open models are fueling progress. FLIP2 introduced public benchmarks for predicting protein mutation effects, while KERMT is a new BioNeMo model for predicting molecular properties important to drug discovery. Synthetic data generation (SDG) drew particular interest, reflecting a shift away from reliance solely on human-labeled data.

NVIDIA's open models have evolved beyond single model releases into a complete research stack: open weights for benchmarking, open datasets for training and adaptation, and open recipes for reasoning, tool use, safety, data curation, and efficient inference. Tools like NeMo Curator and SDG pipelines enable reproducible training data curation at scale.

The ecosystem building on top is substantial. Sakana AI built its Fugu and Fugu-Ultra models directly on Nemotron 3 Ultra. KiloCode integrated Nemotron into its code-routing architecture, reporting token cost reductions of up to 90%. NAVER extended Nemotron architecture for Korean-language AI research, and Together AI is hosting Nemotron models for accessible open inference.

On the robotics front, Humanoid, LG Electronics, NEURA Robotics, and Noble Machines are adopting NVIDIA Isaac GR00T models for industrial humanoid deployments, while 1X, Agility Robots, Boston Dynamics, and others are building next-generation humanoids using Cosmos world models and Isaac Sim.