GR00T 2 at a glance
At CoRL on November 9, 2026, NVIDIA introduced GR00T 2, an open frontier world action model built on Cosmos and trained on robot and human data, including 100,000 hours of egocentric video. In NVIDIA's evaluation it reached 71% average success on 50 tasks with one demonstration each.
What changed with GR00T 2
On November 9, 2026, at the Conference on Robot Learning (CoRL) in Austin, Texas, NVIDIA introduced GR00T 2, an open frontier world action model built on NVIDIA Cosmos. It is trained on diverse robot and human data, including 100,000 hours of NVIDIA's open-source human egocentric data, and is built to teach robots dexterous manipulation: aligning gears, connecting cables and assembling small components. NVIDIA also released Isaac Lab 3.0, its open robot-learning framework, alongside the Isaac Lab-Arena evaluation framework and two new open benchmarks, DexBench and VABAR.
In NVIDIA's own evaluation, GR00T 2 completed 40 of 50 evaluated tasks at least once with no task-specific demonstrations, reached 71% average success across the 50 tasks with one demonstration per task, and 95% with five. As of November 9, 2026, NVIDIA says the model ranks among the top three models across all vision-language-action and world action model categories on the RoboCasa365, RoboLab, RoboDojo and MolmoSpaces benchmarks. AI agents can call the model's API and supporting tools to run skills in sequence and recover from errors. The announcement states no licence, parameter count or weights download, and no GR00T 2 weights were posted on Hugging Face when we checked on November 10, 2026.
GR00T 2 in numbers
Values are copied from the linked source on the date shown. They are not IA Compa measurements.
- Human egocentric training data (NVIDIA)as of · Source for Human egocentric training data (NVIDIA): blogs.nvidia.com (opens in a new tab)
- 100,000 hours
- Average success on 50 tasks, one demonstration per task (NVIDIA)as of · Source for Average success on 50 tasks, one demonstration per task (NVIDIA): blogs.nvidia.com (opens in a new tab)
- 71%
- Average success on 50 tasks, five demonstrations per task (NVIDIA)as of · Source for Average success on 50 tasks, five demonstrations per task (NVIDIA): blogs.nvidia.com (opens in a new tab)
- 95%
Where to compare GR00T 2
Independent leaderboards and official resources for comparing this model yourself. Rankings change often; each site shows its own current data.
Why GR00T 2 matters for businesses
Why it matters: physical AI is moving from locomotion demos to fine manipulation, the kind of work factories and warehouses actually pay for. Foxconn is already applying GR00T 2 to factory automation for server manufacturing, Siemens is evaluating it for industrial use cases and Nexuni is testing it for robotic laundry handling. For automation businesses, an open robot foundation model that learns a new task from a handful of demonstrations lowers the cost of piloting robots on varied, small-batch work.
Get AI News by email
Verified AI stories, each with its official source, on Monday and Thursday mornings. Add breaking news to hear about the biggest ones as soon as we publish them.
Every story links its official source
Free, in English or Spanish
Unsubscribe with one click, any time
Sources for this story
- Practice Makes Precise: At CoRL, NVIDIA and Partners Reveal How Robots Rehearse Delicate Work in Simulation (opens in a new tab)Official
NVIDIA Blog · blogs.nvidia.com
Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.
- #robotics
- #world-action-model
- #open-model
- #physical-ai
- #isaac-lab
- #manipulation
- #cosmos
