GR00T N1.7
Overview
GR00T N1.7 is NVIDIA's vision language action (VLA) model for humanoid robots, released at GTC in March 2026. It is part of the broader NVIDIA Physical AI Stack and represents a major step forward from earlier research prototypes — NVIDIA describes it as "commercially viable for real-world deployment", a phrase that distinguishes it from previous versions of the GR00T series. NVIDIA AI Models 2026 Guide
Key Capabilities
- Full body control for humanoid robots, enabling generalist manipulation and locomotion policies.
- Integrates with Cosmos Reason (part of Cosmos 2.5) for enhanced reasoning and contextual understanding of the physical environment.
- Acts as an open-source model available free on Hugging Face, GitHub, and NVIDIA Build under a commercial-friendly license. NVIDIA AI Models 2026 Guide
Benchmark Performance
GR00T N1.7 currently tops two leading robotics evaluation leaderboards:
No other company has an open-source VLA model competitive with GR00T N1.7 at this scale, according to the source. NVIDIA AI Models 2026 Guide
Architecture & Training
GR00T N1.7 is trained using synthetic data generated by the Cosmos 2.5 world foundation model platform, which can simulate rare edge-case environments and produce large volumes of diverse training scenarios. This simulation-first approach dramatically reduces physical hardware development costs — teams can validate policies in simulation before deploying on real robots. NVIDIA AI Models 2026 Guide
Adoption
Early adopters of GR00T N1.7 at commercial scale include:
| Organization | Use Case |
|---|---|
| LG Electronics | Humanoid robot scaling |
| NEURA Robotics | Humanoid robot scaling |
NVIDIA AI Models 2026 Guide
These are production deployments, not pilot programs. NVIDIA AI Models 2026 Guide
How GR00T N1.7 Fits the NVIDIA AI Stack
GR00T N1.7 is deployed via NVIDIA NIM (Inference Microservices) and discoverable through NVIDIA Build, tying it tightly to NVIDIA's hardware ecosystem — particularly the Rubin Hardware Platform and Blackwell-generation GPUs. NVIDIA AI Models 2026 Guide
Hardware Requirements
GR00T N1.7 is optimized for NVIDIA data center-class GPUs. The broader NVIDIA physical AI strategy intentionally makes model weights free while requiring NVIDIA hardware for production-grade inference — a deliberate hardware lock-in mechanism embedded in a free software release. NVIDIA AI Models 2026 Guide
Competitive Positioning
In the open-source robotics VLA space, GR00T N1.7 has no serious competitors as of early 2026. Alpamayo R1 covers the adjacent autonomous vehicle domain, while Waymo (Google/Alphabet) leads in deployed autonomous driving miles but operates as a closed, proprietary system. NVIDIA AI Models 2026 Guide
| Capability | GR00T N1.7 | Nearest Open-Source Alternative |
|---|---|---|
| Generalist humanoid VLA | ✅ Best-in-class | None comparable |
| MolmoSpaces leaderboard | ✅ #1 | — |
| RoboArena leaderboard | ✅ #1 | — |
| Commercial license | ✅ Free | Varies |
| Simulation integration | ✅ Cosmos 2.5 | Limited |
NVIDIA AI Models 2026 Guide
Roadmap: GR00T N2
Jensen Huang previewed GR00T N2 at GTC March 2026. Key claims:
- Succeeds at new tasks in new environments more than twice as often as competing VLA models.
- Expected availability: end of 2026.
GR00T N2 is not yet released; all performance claims above are from NVIDIA's own preview. NVIDIA AI Models 2026 Guide
Relationship to the Broader NVIDIA Ecosystem
GR00T N1.7 is one pillar of NVIDIA's NVIDIA Physical AI Stack, alongside Cosmos 2.5 (simulation), Alpamayo R1 (autonomous vehicles), and the Nemotron 3 language model family (agentic AI). Together these form NVIDIA's answer to physical-world AI — a domain where neither OpenAI, Anthropic, nor Meta currently offers open-source alternatives. See NVIDIA vs. OpenAI, Google, AMD & Meta: AI Competitive Landscape 2026 for a full comparison. NVIDIA AI Models 2026 Guide
All NVIDIA open models, including GR00T N1.7, are part of the NVIDIA Open Model Ecosystem and are accessible through NVIDIA Build.