π₀.₅ (Pi05) is a Vision-Language-Action model from Physical Intelligence designed for open-world generalization: it evolves π₀ to generalize to entirely new environments and situations that were never seen during training. The LeRobot implementation is adapted from their open-source OpenPI repository. This policy has been trained and pushed to the Hub using LeRobot. Learn how to train and run it in the LeRobot pi05 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…
Vision-language-action (VLA) models have become a dominant paradigm for generalist embodied agents, demonstrating strong complex and long-horizon task completion in structured settings.
Parameters4.1B
Context—
Weights33.0 GB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads6k
Runs On
What it takes to serve GigaBrain-0.7-3.5B-Base (4.1B 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 | 8.2 GB | 9.9 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 4.1 GB | 4.9 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 2.1 GB | 2.5 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x 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, read Sep 18, 2026.
Model Card
By GigaAI, published under apache-2.0, revision 2936d6f26327.
Vision-language-action (VLA) models have become a dominant paradigm for generalist embodied agents, demonstrating strong complex and long-horizon task completion in structured settings. Yet it remains an open question whether current VLA systems can benefit from more effective architectural design, scale to substantially larger and more heterogeneous data regimes, and achieve broader generalization across tasks and embodiments. To this end, we present GigaBrain-0.7, an embodied foundation model with substantially improved generalization across diverse robot embodiments. Specifically, GigaBrain-0.7 unifies understanding, prediction, and action through a three-system architecture, scales…
Read GigaAI's full model card
GigaBrain-0.7: Scaling Embodied Foundation Models to Emergent Capabilities with a Three-System Architecture
[](https://opensource.org/licenses/Apache-2.0) [](https://gigaai.cc/blog/gigabrain07) [](https://arxiv.org/abs/2608.15875) [](https://github.com/open-gigaai/giga-brain-0/blob/main/tech_report/GigaBrain-0.7.pdf) [](https://huggingface.co/open-gigaai/GigaBrain-0.7-3.5B-Base) [](https://huggingface.co/datasets/open-gigaai/GigaBrain-0.7-SampleData)Introduction
Vision-language-action (VLA) models have become a dominant paradigm for generalist embodied agents, demonstrating strong complex and long-horizon task completion in structured settings. Yet it remains an open question whether current VLA systems can benefit from more effective architectural design, scale to substantially larger and more heterogeneous data regimes, and achieve broader generalization across tasks and embodiments. To this end, we present GigaBrain-0.7, an embodied foundation model with substantially improved generalization across diverse robot embodiments. Specifically, GigaBrain-0.7 unifies understanding, prediction, and action through a three-system architecture, scales pretraining to over 37,000 hours of heterogeneous embodied data, and introduces one-stage alignment training that jointly optimizes vision-language understanding and multi-embodiment action generation. Compared with the preceding GigaBrain-0 series and prior state-of-the-art models including π0.5, GigaBrain-0.7 achieves substantial improvements in foundation zero-shot capabilities, language-conditioned instruction following, and post-training task success rates. In particular, on our in-house Maker H01 platform and mainstream robot embodiments, GigaBrain-0.7 demonstrates strong task adaptability and completion ability across both home and industrial scenarios.
Citation
@article{gigabrainteam2026gigabrain07,
title={GigaBrain-0.7: Scaling Embodied Foundation Models to Emergent
Capabilities with a Three-System Architecture},
author={GigaBrain Team and others},
journal={arXiv preprint arXiv:2608.15875},
year={2026},
eprint={2608.15875},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2608.15875},
}
Identity and Version
- Repository
- open-gigaai/GigaBrain-0.7-3.5B-Base
- Publisher
- GigaAI
- Task
- Robotics
- Modality
- Control
- Library
- diffusers
- Parameters
- 4.1B parameters
- Languages
- en
- Revision
- 2936d6f263279a58d4f9153b2efa597812c194cb
- First published
- 2026-08-24
- Last updated
- 2026-08-25
Files and Weights
10 files, 33.0 GB in total. The weights are 5 files totalling 33.0 GB in bin, safetensors.
Weights5 files · 33.0 GB
Configuration2 files · 137.6 KB
Documentation1 file · 3.0 KB
Other1 file · 2.3 MB
Repository1 file · 1.6 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| diffusion_pytorch_model-00001-of-00004.safetensors | Weights | 5.0 GB | 223bda49a0a0 |
| diffusion_pytorch_model-00002-of-00004.safetensors | Weights | 4.9 GB | 9fb0122adfd0 |
| diffusion_pytorch_model-00003-of-00004.safetensors | Weights | 4.2 GB | 7db9419afed9 |
| diffusion_pytorch_model-00004-of-00004.safetensors | Weights | 2.4 GB | 02506a600cbf |
| diffusion_pytorch_model.bin | Weights | 16.5 GB | d8270bd58cf1 |
| config.json | Configuration | 1.9 KB | — |
| diffusion_pytorch_model.safetensors.index.json | Configuration | 135.7 KB | — |
| README.md | Documentation | 3.0 KB | — |
| gigabrain07_arch.png | Other | 2.3 MB | 39da63661083 |
| .gitattributes | Repository | 1.6 KB | — |
License and Download
- License
- apache-2.0
- Access
- Open weights, no gate
- Download size
- 33.0 GB
Download from GigaAI
Released by GigaAI through its official repository on Hugging Face. Read the license.
Built From
- Described by arXiv:2608.15875
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 33.0 GB |
| 16-bit | 8.2 GB |
| 8-bit | 4.1 GB |
| 4-bit | 2.1 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About GigaBrain-0.7-3.5B-Base
How much GPU memory does GigaBrain-0.7-3.5B-Base need?
About 9.9 GB at 16-bit and 2.5 GB at 4-bit: the weights (4.1B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run GigaBrain-0.7-3.5B-Base 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.
Can I use GigaBrain-0.7-3.5B-Base commercially?
Yes. GigaBrain-0.7-3.5B-Base is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.
Similar Models
Final checkpoint after 60,000 optimization steps. This is a trained policy, not an evaluation result. Policy weights, policy configuration, preprocessing/postprocessing and normalization states are at the repository root. Training resume files are in trainingstate/. Host-specific paths were removed from JSON metadata; supply local dataset/output paths when resuming. The tokenizer reference points to google/paligemma-3b-pt-224 (training revision 35e4f46485b4d07967e7e9935bc3786aad50687c). For subtask models, supply the corresponding per-frame subtask as the policy task text. For 3-view models also supply the dataset-defined keyframe image. No real-robot evaluation metrics are claimed here.…
Final checkpoint after 60,000 optimization steps. This is a trained policy, not an evaluation result. Policy weights, policy configuration, preprocessing/postprocessing and normalization states are at the repository root. Training resume files are in trainingstate/. Host-specific paths were removed from JSON metadata; supply local dataset/output paths when resuming. The tokenizer reference points to google/paligemma-3b-pt-224 (training revision 35e4f46485b4d07967e7e9935bc3786aad50687c). For subtask models, supply the corresponding per-frame subtask as the policy task text. For 3-view models also supply the dataset-defined keyframe image. No real-robot evaluation metrics are claimed here.…
π₀.₅ (Pi05) is a Vision-Language-Action model from Physical Intelligence designed for open-world generalization: it evolves π₀ to generalize to entirely new environments and situations that were never seen during training. The LeRobot implementation is adapted from their open-source OpenPI repository. This policy has been trained and pushed to the Hub using LeRobot. Learn how to train and run it in the LeRobot pi05 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…
π₀.₅ is a Vision-Language-Action (VLA) model with open-world generalization from Physical Intelligence, co-trained on robot demonstrations and large-scale multimodal data to execute long-horizon tasks in unseen real-world environments. Checkpoint trained and evaluated on LIBERO tasks Note: This model currently supports only the flow-matching action head for π₀.₅ training and inference. Other components from the original work (e.g., subtask prediction, action tokenization, or RL) were not released upstream and are not included here, though the LeRobot team is actively working to support them. Original paper: π0.5: A Vision-Language-Action Model with Open-World Generalization For full…
π₀.₅ is a Vision-Language-Action (VLA) model with open-world generalization from Physical Intelligence, co-trained on robot demonstrations and large-scale multimodal data to execute long-horizon tasks in unseen real-world environments. Note: This model currently supports only the flow-matching action head for π₀.₅ training and inference. Other components from the original work (e.g., subtask prediction, action tokenization, or RL) were not released upstream and are not included here, though the LeRobot team is actively working to support them. Original paper: π0.5: A Vision-Language-Action Model with Open-World Generalization For full installation details (including optional video…