NVIDIA's open models and infrastructure have become foundational to AI research, with adoption spanning robotics, life s

The International Conference on Machine Learning 2026 demonstrated a clear shift toward open frontier models and open AI infrastructure as the basis for modern AI research. NVIDIA had 74 papers accepted at the conference, with approximately 2,000 accepted papers citing NVIDIA GPUs and 145 citing NVIDIA Nemotron, an open model family with accompanying datasets. Hundreds of additional papers build on other NVIDIA open model families including Cosmos, Isaac GR00T, BioNeMo and others, spanning physical AI, robotics, autonomous vehicles and biomedical research.

Vision and video generation, reinforcement learning for large language models and agent training remained prominent research themes this year, along with AI inference, while several new areas emerged. Robot world models drew significant attention, exemplified by DreamDojo, which learns physical behavior from human video using NVIDIA Cosmos models to predict how robots handle objects and operate in unfamiliar environments. This enables researchers to evaluate policies and test actions virtually, accelerating development without physical deployment costs and risks.

AI for life sciences research was enabled by NVIDIA BioNeMo open models and research contributions. Papers like FLIP2 introduce public benchmarks for predicting the effects of protein mutations, while KERMT, a new BioNeMo model, predicts molecular properties important to drug discovery. Synthetic data generation drew particular interest at ICML, with Nemotron and physical AI datasets reflecting a broader shift toward training at scale without relying solely on human-labeled data.

The research shows Nemotron functioning as a complete research stack with open weights, open datasets for training and adaptation, and open recipes for reasoning, tool use, safety, data curation and inference. NeMo Curator and its supported datasets provide researchers with reproducible foundations for training data curation, while synthetic data generation tools enable creating high-quality training sets at scales previously impractical. Cosmos 3, an open omnimodel family, offers a generational advance for building robots, autonomous vehicles and vision AI. Additional model families including Alpamayo for autonomous vehicle development, Isaac GR00T for robotics and BioNeMo for biomedical applications accelerate research across industries.

This momentum extends beyond NVIDIA's own work. Basecamp Research developed EDEN, a DNA foundation model for interpreting and designing genetic sequences. Merck and Company uses KERMT to predict drug molecule behavior and efficacy. Sakana AI built its Fugu models directly on Nemotron 3 Ultra for AI research automation. KiloCode integrated Nemotron into its code-routing architecture, achieving token cost reductions of up to 90 percent. NAVER developed its own model using the Nemotron architecture for Korean-language AI research. Together AI hosts Nemotron models on its platform for accessible researcher access. Robotics companies including Humanoid, LG Electronics, NEURA Robotics and Noble Machines are adopting Isaac GR00T models for industrial humanoid deployment, while 1X, Agility, Agile Robots, Boston Dynamics, Hexagon Robotics and Mentee are building next-generation humanoids using Cosmos world models, Isaac Sim and Isaac Lab for accelerated development and validation.