# Nvidia's GR00T 2 Robot Model Claims 71% Success From One Demonstration, on the Same Day Anthropic Claimed a Robotics Benchmark Lead

Times of AI Desk · 2026-11-10 · Research

[https://timesof.ai/2026/11/nvidia-gr00t-2-world-action-model-71-percent-one-demo-corl](https://timesof.ai/2026/11/nvidia-gr00t-2-world-action-model-71-percent-one-demo-corl)

> Nvidia introduced GR00T 2 at the Conference on Robot Learning on November 9, an open world action model built on its Cosmos models and trained on robot and human data including 100,000 hours of human egocentric video. Nvidia says it completed 40 of 50 tasks at least once with no task-specific demonstrations and averaged 71% success with one demonstration per task, 95% with five. Foxconn and Siemens are testing it. The figures are Nvidia's own and use different setups from other labs' robotics claims.

Robotics had two competing claims on the same Monday. Nvidia says its purpose-built robot model now ranks near the top of the field's leaderboards, while Anthropic said its general-purpose model beat most systems built specifically to control robots. Both are self-reported, and they do not measure the same thing.

## What Nvidia released

At the **Conference on Robot Learning (CoRL)** in Austin, Nvidia introduced **GR00T 2** on **November 9**, describing it as "an open frontier world action model built on NVIDIA Cosmos." A world action model predicts how the scene will change along with the robot's next actions. Nvidia says it trained GR00T 2 on diverse robot and human data, including 100,000 hours of its open-source human egocentric data.

Nvidia's headline figures:

| Claim (Nvidia's) | Figure |
| --- | --- |
| Tasks completed at least once with no task-specific demonstrations | 40 of 50 |
| Average success across 50 tasks, one demonstration per task | 71% |
| Average success across 50 tasks, five demonstrations per task | 95% |

Nvidia also says GR00T 2 can transfer skills from human demonstrations to different robot bodies without robot demonstrations, and that AI agents can call its API to run skills in sequence and recover from errors. It says the model ranks among the top three across vision-language-action and world action model categories on the RoboCasa365, RoboLab, RoboDojo and MolmoSpaces benchmarks.

Nexuni is testing it for laundry handling, Foxconn is applying it to server-manufacturing automation and Siemens is evaluating it, according to Nvidia. The Robot Report carried the same figures, attributed to Nvidia.

## The training stack around it

Nvidia also released **Isaac Lab 3.0**, its open-source robot-learning framework, which it says achieved up to 2.5 times the training throughput of Isaac Lab 2.3 in a vision-based dexterous manipulation benchmark when paired with the Newton physics engine. In one demonstration, a single policy learned all 20 tasks on the NIST Assembly Task Board with more than 98% training success, against 81% for separate per-task policies, using roughly 39 times less GPU time, Nvidia says.

Nvidia also released an open workflow that turns human video into simulated robot demonstrations. Collecting that kind of data has drawn investors too, as in [GenRobot's $100 million round](https://timesof.ai/2026/11/genrobot-ai-100m-series-a-plus-robot-training-data) earlier this month.

## Claims vs checks

Nvidia's ranking claim is a top-three placement, not a lead, and it did not publish the scores alongside its announcement. Anthropic said the same day that Claude Fable 5.2 scored 64.9% on RoboDojo Sim at maximum effort and outperformed "most AI systems built specifically to control robots." Anthropic's figure comes from its own internal runs on its own harness, so the two claims cannot be placed side by side.

## Limits

- Every performance figure here is Nvidia's own; we did not find an independent evaluation of GR00T 2.
- Nvidia calls the model open but its announcement did not state licence terms or where weights are published; we did not check model repositories.
- Partner deployments are described as testing or evaluation, not production use.

## Sources

- [Nvidia: Practice Makes Precise: At CoRL, NVIDIA and Partners Reveal How Robots Rehearse Delicate Work in Simulation (November 9, 2026)](https://blogs.nvidia.com/blog/robotics-simulation-advancements-corl/)
- [The Robot Report: NVIDIA launches GR00T 2 and Isaac Lab 3.0 at CoRL (November 9, 2026)](https://www.therobotreport.com/nvidia-launches-gr00t-2-isaac-lab-3-0-corl/)
- [Anthropic: Claude Fable 5.2 and Mythos 5.2 (November 9, 2026)](https://www.anthropic.com/claude-fable-and-mythos-5-2)