# GR00T-N1.7-3B by NVIDIA: Open-Weight Model
Source: https://savrn.com/models/gr00t-n1-7-3b
Markdown alternate of the page above; the site index is https://savrn.com/llms.txt

---

## Runs On

What it takes to serve GR00T-N1.7-3B (3.1B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.

| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
| --- | --- | --- | --- | --- | --- |
| 16-bit | 6.3 GB | 7.5 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 |
| 8-bit | 3.1 GB | 3.8 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 |
| 4-bit | 1.6 GB | 1.9 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 |

Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the [SAVRN Index](https://savrn.com/ai-index/pricing/gpus), read Oct 24, 2026.

[GR00T-N1.7-3B on every accelerator the SAVRN Index prices, at every precision](https://savrn.com/models/gr00t-n1-7-3b/gpus)

## SAVRN's Notes on GR00T-N1.7-3B

GR00T N1.7 is NVIDIA's open foundation model for humanoid robots. It takes language and camera images and outputs actions for manipulation tasks, and it is designed to transfer across different robot bodies. At 3.1B parameters it can run on robot-class onboard compute.

Our record does not show a license for this repository, so read NVIDIA's terms on the card before any commercial use. For robotics teams it is a starting policy to fine-tune with your own demonstrations, not a finished controller.

SAVRN Research, 2026-09-18

## Model Card

NVIDIA Isaac GR00T N1.7 is an open foundation model for generalized humanoid robot reasoning and skills. This cross-embodiment model takes multimodal input, including language and images, to perform manipulation tasks in diverse environments. Developers and researchers can post-train GR00T N1.7 with real or synthetic data for their specific humanoid robot or task. Isaac GR00T N1.7 is the medium-sized version of our model built using pre-trained vision and language encoders, and uses a flow matching action transformer to model a chunk of actions conditioned on vision, language and proprioception. A detailed description of the Isaac GR00T N1.X architecture is provided in the GROOT N1 White…

Excerpt from the card by NVIDIA.

## Configuration

Architecture

Gr00tN1d7

Hidden size

1,024

Model type

Gr00tN1d7

## Identity and Version

Repository

nvidia/GR00T-N1.7-3B

Publisher

NVIDIA

Task

Robotics

Modality

Control

Library

Not stated by the source

Parameters

3.1B parameters

Languages

Not stated by the source

Revision

2fc962b973bccdd5d8ce4f67cc63b264d6886495

First published

2026-02-25

Last updated

2026-04-23

## Files and Weights

27 files, 6.9 GB in total. The weights are 5 files totalling 6.9 GB in bin, npz, pt, safetensors.

Weights5 files · 6.9 GB

Configuration13 files · 12.1 MB

Documentation5 files · 21.4 KB

Other3 files · 3.2 MB

Repository1 file · 1.6 KB

Every file

| File | Type | Size | SHA-256 |
| --- | --- | --- | --- |
| experiment_cfg/initial_actions.npz | Weights | 1.6 MB | c34d9e522cf1 |
| model-00001-of-00002.safetensors | Weights | 5.0 GB | 8a1a1d8a33c9 |
| model-00002-of-00002.safetensors | Weights | 1.9 GB | c3f61940deb2 |
| scheduler.pt | Weights | 1.3 KB | 5ae78127cbbf |
| training_args.bin | Weights | 8.3 KB | 3c45bcc6f491 |
| config.json | Configuration | 2.1 KB | — |
| embodiment_id.json | Configuration | 2.1 KB | — |
| experiment_cfg/conf.yaml | Configuration | 31.2 KB | — |
| experiment_cfg/config.yaml | Configuration | 35.0 KB | — |
| experiment_cfg/dataset_statistics.json | Configuration | 4.3 MB | — |
| experiment_cfg/final_model_config.json | Configuration | 1.5 KB | — |
| experiment_cfg/final_processor_config.json | Configuration | 43.4 KB | — |
| model.safetensors.index.json | Configuration | 105.0 KB | — |
| processor_config.json | Configuration | 21.1 KB | — |
| statistics.json | Configuration | 4.3 MB | — |
| trainer_state.json | Configuration | 3.2 MB | — |
| wandb_config.json | Configuration | 200 B | — |
| zero_to_fp32.py | Configuration | 33.3 KB | — |
| EXPLAINABILITY.md | Documentation | 2.5 KB | — |
| LICENSE | Documentation | 4.1 KB | — |
| PRIVACY.md | Documentation | 551 B | — |
| README.md | Documentation | 13.1 KB | — |
| SAFETY_and_SECURITY.md | Documentation | 1.1 KB | — |
| SUCCESS | Other |  | — |
| latest | Other | 17 B | — |
| model-architecture.png | Other | 3.2 MB | bbf161c66daa |
| .gitattributes | Repository | 1.6 KB | — |

## License and Download

License

Not stated by the source

Access

Open weights, no gate

Download size

6.9 GB

[Download from NVIDIA](https://huggingface.co/nvidia/GR00T-N1.7-3B)

Released by NVIDIA through its official repository on Hugging Face.

## Built From

- Described by [arXiv:2503.14734](https://savrn.com/papers/gr00t-n1-an-open-foundation-model-for-generalist-humanoid-robots)

## Memory Requirements

| Precision | Weights in memory |
| --- | --- |
| As published | 6.9 GB |
| 16-bit | 6.3 GB |
| 8-bit | 3.1 GB |
| 4-bit | 1.6 GB |

Weights only, from the published parameter count; the key-value cache and runtime add to this.

## Built on This Model

- Derived from[gr00t-n17-bigenlight-50per-15hz-eef-observed-delta-bc10k-headonly-20260926](https://savrn.com/models/gr00t-n17-bigenlight-50per-15hz-eef-observed-delta-bc10k-headonly-20260926)
- Derived from[GR00T-N1.7-ForceBenchmark-MoE-vasev6-Transformer-30K-20260926](https://savrn.com/models/gr00t-n1-7-forcebenchmark-moe-vasev6-transformer-30k-20260926)
- Derived from[GR00T-N1.7-ForceBenchmark-MoE-usb-Transformer-30K-20260926](https://savrn.com/models/gr00t-n1-7-forcebenchmark-moe-usb-transformer-30k-20260926)
- Derived from[GR00T-N1.7-ForceBenchmark-MoE-whiteboardv9-Transformer-30K-20260926](https://savrn.com/models/gr00t-n1-7-forcebenchmark-moe-whiteboardv9-transformer-30k-20260926)
- Derived from[GR00T-N1.7-ForceBenchmark-MoE-usb-MLP-30K-20260926](https://savrn.com/models/gr00t-n1-7-forcebenchmark-moe-usb-mlp-30k-20260926)
- Derived from[GR00T-N1.7-ForceBenchmark-MoE-vasev6-MLP-30K-20260926](https://savrn.com/models/gr00t-n1-7-forcebenchmark-moe-vasev6-mlp-30k-20260926)
- Derived from[GR00T-N1.7-ForceBenchmark-MoE-whiteboardv9-MLP-30K-20260926](https://savrn.com/models/gr00t-n1-7-forcebenchmark-moe-whiteboardv9-mlp-30k-20260926)
- Derived from[GR00T-N1.7-ForceBenchmark-MoE-weightsorting-MLP-30K-20260926](https://savrn.com/models/gr00t-n1-7-forcebenchmark-moe-weightsorting-mlp-30k-20260926)
- Derived from[GR00T-N1.7-ForceBenchmark-Force8-PickCubeV4-Obs2-H10-30K-20260928](https://savrn.com/models/gr00t-n1-7-forcebenchmark-force8-pickcubev4-obs2-h10-30k-20260928)
- Derived from[GR00T-ForceBenchmark-WeightSortingV6-PickCubeV4-5x30K-20260929](https://savrn.com/models/gr00t-forcebenchmark-weightsortingv6-pickcubev4-5x30k-20260929)
- Derived from[GR00T-N1.7-ForceBenchmark-USBv4-Q02-Q09-MoE-NoForce-9x30K-20261011](https://savrn.com/models/gr00t-n1-7-forcebenchmark-usbv4-q02-q09-moe-noforce-9x30k-20261011)

## Questions About GR00T-N1.7-3B

### How much GPU memory does GR00T-N1.7-3B need?

About 7.5 GB at 16-bit and 1.9 GB at 4-bit: the weights (3.1B parameters) plus a working margin. A long context needs more.

### What is the cheapest GPU to run GR00T-N1.7-3B on?

At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

## Similar Models

Model · Robotics

### [GR00T-N1.7-DROID](https://savrn.com/models/gr00t-n1-7-droid)

[NVIDIA](https://savrn.com/model-publishers/nvidia)

NVIDIA Isaac GR00T N1.7 is an open foundation model for generalized humanoid robot reasoning and skills. This cross-embodiment model takes multimodal input, including language and images, to perform manipulation tasks in diverse environments. Developers and researchers can post-train GR00T N1.7 with real or synthetic data for their specific humanoid robot or task. Isaac GR00T N1.7 is the medium-sized version of our model built using pre-trained vision and language encoders, and uses a flow matching action transformer to model a chunk of actions conditioned on vision, language and proprioception. A detailed description of the Isaac GR00T N1.X architecture is provided in the GROOT N1 White…

Open weights 3.1B parameters

[View model](https://savrn.com/models/gr00t-n1-7-droid)

Model · Robotics

### [so101_vials_dr_randomized_grasp_sim250_GR00T17_real10_ft5000_20261008](https://savrn.com/models/so101-vials-dr-randomized-grasp-sim250-gr00t17-real10-ft5000-20261008)

[Lior Ben Horin](https://savrn.com/model-publishers/liorbenhorin-nv)

Independent further-training of the new sim-only 20,000-step policy on 10 real10 episodes for 5,000 additional steps. Both real10 and Cosmos10 branches initialize from the same sim-only parent; neither continues from the other branch. - Native parent processors and statistics retained, with stock clipping; no unclipping ablation. - Visual tuning enabled, language tuning disabled, BF16 autocast with FP32 parameters; image augmentation enabled. - Checkpoints at 2,500 and 5,000 additional steps. This repository contains the final 5,000-step continuation. The native policy and processor files must remain together. See training-provenance.json for parent weight hash and exact provenance.…

Open weights apache-2.0 3.1B parameters lerobot

[View model](https://savrn.com/models/so101-vials-dr-randomized-grasp-sim250-gr00t17-real10-ft5000-20261008)

Model · Robotics

### [so101_cvs2_80ep_groot17](https://savrn.com/models/so101-cvs2-80ep-groot17)

[Okada Yudai](https://savrn.com/model-publishers/yudai0209)

GR00T N1.7 is an open, cross-embodiment foundation model from NVIDIA for generalized humanoid robot reasoning and skills. It uses a Cosmos-Reason2/Qwen3-VL backbone and a flow-matching action transformer to predict actions conditioned on vision, language, and proprioception. This policy has been trained and pushed to the Hub using LeRobot. Learn how to train and run it in the LeRobot groot guide, or browse the full documentation. The policy consumes these observation features and produces these action features. Inputs Outputs New to LeRobot? These guides cover the full workflow: - Install LeRobot — set up the lerobot package. - Hardware setup — assemble, wire, and calibrate your robot and…

Open weights apache-2.0 3.1B parameters lerobot

[View model](https://savrn.com/models/so101-cvs2-80ep-groot17)

Model · Robotics

### [groot-n17-droid-pour-water-10ep](https://savrn.com/models/groot-n17-droid-pour-water-10ep)

[Sang Nguyen](https://savrn.com/model-publishers/tsangb34)

GR00T N1.7 is an open, cross-embodiment foundation model from NVIDIA for generalized humanoid robot reasoning and skills. It uses a Cosmos-Reason2/Qwen3-VL backbone and a flow-matching action transformer to predict actions conditioned on vision, language, and proprioception. This policy has been trained and pushed to the Hub using LeRobot. Learn how to train and run it in the LeRobot groot guide, or browse the full documentation. The policy consumes these observation features and produces these action features. Inputs Outputs New to LeRobot? These guides cover the full workflow: - Install LeRobot — set up the lerobot package. - Hardware setup — assemble, wire, and calibrate your robot and…

Open weights apache-2.0 3.1B parameters lerobot

[View model](https://savrn.com/models/groot-n17-droid-pour-water-10ep)

Model · Robotics

### [GR00T-N1.7-ForceBenchmark-MoE-usb-MLP-30K-20260926](https://savrn.com/models/gr00t-n1-7-forcebenchmark-moe-usb-mlp-30k-20260926)

[Linqi Han](https://savrn.com/model-publishers/hanlinqi)

This repository contains the model-only checkpoint saved at 30,000 training steps. The files at the repository root are the original contents of checkpoint-30000. Full model and processor settings are in config.json, processorconfig.json, and experimentcfg/. The training run did not include an evaluation. Source checkpoint: outputs/moeobs2h10fourtasks20260926191812/pair67/usb/moe/checkpoint-30000.

Open weights 3.2B parameters

[View model](https://savrn.com/models/gr00t-n1-7-forcebenchmark-moe-usb-mlp-30k-20260926)

Model · Robotics

### [GR00T-N1.7-ForceBenchmark-MoE-vasev6-MLP-30K-20260926](https://savrn.com/models/gr00t-n1-7-forcebenchmark-moe-vasev6-mlp-30k-20260926)

[Linqi Han](https://savrn.com/model-publishers/hanlinqi)

This repository contains the model-only checkpoint saved at 30,000 training steps. The files at the repository root are the original contents of checkpoint-30000. Full model and processor settings are in config.json, processorconfig.json, and experimentcfg/. The training run did not include an evaluation. Source checkpoint: outputs/moeobs2h10fourtasks20260926191812/pair67/vasev6/moe/checkpoint-30000.

Open weights 3.2B parameters

[View model](https://savrn.com/models/gr00t-n1-7-forcebenchmark-moe-vasev6-mlp-30k-20260926)

## NVIDIA

[All models and datasets](https://savrn.com/model-publishers/nvidia)

## Versions

- [2fc962b973bc](https://savrn.com/models/gr00t-n1-7-3b/versions/2fc962b973bc) · current 2026-09-18

## Explore More

- [All robotics models](https://savrn.com/models/tasks/robotics)
- [Model comparisons](https://savrn.com/models/comparisons)
- [The model directory](https://savrn.com/models)
- [Open model prices by host](https://savrn.com/ai-index/pricing/open-models)

## Source

- Repository metadata, read 2026-10-20.
- [NVIDIA website](https://build.nvidia.com/models)
- [Hugging Face record](https://huggingface.co/nvidia/GR00T-N1.7-3B)
- [How the hub is built](https://savrn.com/model-hub/methodology)