Big techSanta Clara, CA, United StatesFounded 1993
Publishes the open Isaac GR00T N robot foundation models and the Cosmos world foundation models for physical AI.
NVIDIA, founded on April 5, 1993 by Jensen Huang, Chris Malachowsky and Curtis Priem and headquartered in Santa Clara, California, is a publicly listed maker of GPUs and AI computing platforms. Its robotics work includes the Isaac GR00T platform; in March 2025 it released Isaac GR00T N1, an open, customizable foundation model for humanoid robots. NVIDIA Research's GEAR (Generalist Embodied Agent Research) lab, led by Jim Fan and Yuke Zhu, builds foundation models for embodied agents, including GR00T. In September 2026 NVIDIA signed a definitive agreement to acquire Hugging Face, expected to close in the first half of 2027.
Sources: nvidia.com (17 Oct 2026) · nvidianews.nvidia.com (18 Mar 2025) · research.nvidia.com (17 Oct 2026) · sec.gov (2 Sep 2026)
Corporate division · Robotics/GR00T work is done by groups within NVIDIA Corporation (NASDAQ: NVDA); no separate funding.
#1 of 104 model companies · score 95.9 / 100 · updated 19 Oct 2026
Each bar shows an input from 0 to 100% before weighting. 8 of 8 listed models have open weights. Most-starred official repository: NVIDIA/cosmos (12.1k stars, 19 Oct 2026).
The score (0–100) measures public, checkable evidence about a company's robot models. It does not measure private capability. A company that publishes no benchmark results and no weights can still have strong models. Every input is computed from the data on each company page, and every number there links to its source.
| Input | Weight | How it is measured |
|---|---|---|
| Published benchmarks | 35% | For each scored benchmark, the company's best result divided by the best result any company has published, averaged over the benchmarks it reports. No published results scores 0. Results come only from the model's own paper, model card or the official benchmark leaderboard. Scored: LIBERO (4-suite average success, %); SimplerEnv Google Robot, visual matching (%); SimplerEnv Google Robot, variant aggregation (%); SimplerEnv WidowX / Bridge (%); CALVIN ABC→D (average length, 0–5); RoboArena (pairwise rating). |
| Open weights | 20% | Share of the company's listed models whose weights the public can download. |
| Adoption | 20% | GitHub stars on the company's most-starred official model repository, on a log scale relative to the most-starred repository in the list. Hugging Face downloads are shown on company pages but not scored, because gated repositories report zero. |
| Released models | 15% | Number of publicly announced, named models, capped at 5. |
| Disclosed funding | 10% | The sum of the funding rounds listed on the company's profile that were disclosed in USD, on a log scale from $10M to the largest sum in the list. Rounds reported only in other currencies are listed but not converted or counted. This input applies only to private startups and robot makers. For big-tech divisions, research labs and listed companies it is n/a, and its weight is shared across the other inputs in proportion. |
Sources. arXiv papers and official model cards for benchmarks and parameters; official leaderboards (RoboArena); the GitHub and Hugging Face APIs for stars and downloads; company announcements and established press for funding. Missing values are shown as n/a, never estimated.
Updated 19 Oct 2026. Benchmark protocols differ between papers, for example in fine-tuning and number of trials, so small gaps between close scores are not meaningful.
Open weightsWorld model
Mixture-of-transformers omni-model that unifies world generation, physical reasoning and action prediction, replacing the separate Predict/Reason/Transfer families; Super and Nano at launch, Edge later.
No benchmark results on a shared public benchmark have been published for this model.
Open weightsVLA
Commercially licensed VLA with a Cosmos-Reason2-2B (Qwen3-VL-based) backbone and relative end-effector actions shared across human and robot data; launched in early access at GTC 2026.
| Benchmark | Result | Source |
|---|---|---|
| LIBEROAverage success rate across LIBERO-Spatial/Object/Goal/10 (%) | 96.99% | github.comIsaac-GR00T LIBERO README (nvidia/GR00T-N1.7-LIBERO, per-suite fine-tuned): Spatial 97.65, Object 98.45, Goal 97.5, 10 94.35; mean of the 4 suites computed by us (README gives no average). As of 17 Oct 2026 |
| OtherSimplerEnv WidowX/Bridge, average over the repo's 7-task set incl. drawer tasks (%) — not the standard 4-task Bridge setShown, not scored | 62.3% | github.comIsaac-GR00T SimplerEnv README, nvidia/GR00T-N1.7-SimplerEnv-Bridge. As of 17 Oct 2026 |
| OtherSimplerEnv Google Robot (Fractal), average over the repo's 6-task set (%) — VM/VA split not statedShown, not scored | 72.5% | github.comIsaac-GR00T SimplerEnv README, nvidia/GR00T-N1.7-SimplerEnv-Fractal. As of 17 Oct 2026 |
Open weightsEmbodied reasoning
Qwen3-VL-based reasoning VLM for physical common sense and embodied reasoning; the 2B model serves as the GR00T N1.7 backbone.
No benchmark results on a shared public benchmark have been published for this model.
Open weightsWorld model
Video world model for synthetic data generation and simulated policy evaluation, released alongside Cosmos-Transfer2.5.
No benchmark results on a shared public benchmark have been published for this model.
Open weightsVLA
Reasoning VLA for humanoids using a Cosmos-Reason-based VLM and a larger diffusion transformer; announced at CoRL 2025 and listed as available on Hugging Face by January 2026.
| Benchmark | Result | Source |
|---|---|---|
| OtherSimplerEnv WidowX/Bridge, average over the repo's 7-task set incl. drawer tasks (%) — not the standard 4-task Bridge setShown, not scored | 56.6% | github.comIsaac-GR00T SimplerEnv README, nvidia/GR00T-N1.6-bridge fine-tuned checkpoint. As of 17 Oct 2026 |
| OtherSimplerEnv Google Robot (Fractal), average over the repo's 6-task set (%) — VM/VA split not statedShown, not scored | 52% | github.comIsaac-GR00T SimplerEnv README, nvidia/GR00T-N1.6-fractal fine-tuned checkpoint. As of 17 Oct 2026 |
Open weightsVLA
Update to GR00T N1 with a frozen, grounding-improved VLM and a simplified adapter; trained partly on GR00T-Dreams synthetic data.
| Benchmark | Result | Source |
|---|---|---|
| RoboCasaAverage success rate, 30 demos per task (%)Shown, not scored | 47.5% | research.nvidia.comGR00T N1.5 research page; N1 scored 17.4 in the same comparison. As of 11 Jun 2025 |
Open weightsVLA
Dual-system VLA (Eagle-2 VLM plus diffusion-transformer action head) for humanoids, trained on real, human-video and synthetic data.
No benchmark results on a shared public benchmark have been published for this model.
Open weightsWorld model
First Cosmos world foundation models (diffusion and autoregressive video world models plus tokenizers) for generating physical-AI training data; repo now in limited maintenance.
No benchmark results on a shared public benchmark have been published for this model.
