modelhub.help
← All verticals
Embodied AI · Vertical LLM

Embodied AI LLMs

VLA, world models and dexterous manipulation — embodied AI enters open scenarios.

Embodied AI is the fastest-growing track of 2025–2026 — and unlike general LLM APIs, most entries here are research or open-weight models you run on robot hardware rather than call per token. We cover representative VLA base models, world models, dual-arm and high-dexterity manipulation, and flag open-source degree and whole-unit export status to help robot makers and researchers assess usability. Every spec below links to its primary source where one is published.

VLA foundation models

7 models

AgiBot GO-1

智元机器人Fully open source

Generalist embodied foundation model on the ViLLA architecture.

Data: Weights downloadable / on-deviceOverseas: Partial (units exported)
Candidate (候选)Details

UnifoLM-VLA

宇树科技BY-NC-SA 4.0

Open humanoid-robot VLA model.

Data: Hardware-boundOverseas: Yes (units sold globally)
Candidate (候选)Details

InternVLA-M1

上海人工智能实验室Open weights

General vision-language-action VLA research model.

Data: Mainland ChinaOverseas: None
Candidate (候选)Details

GOVLA / NeuroVLA

智平方 AI²Commercial

Brain-inspired embodied VLA model.

Data: Mainland ChinaOverseas: None
Candidate (候选)Details

X-VLA

自变量 X SquareMIT

0.9B fully open-source clothes-folding VLA.

Data: Self-hostableOverseas: None
Candidate (候选)Details

Star1

星动纪元Commercial / research

Humanoid robot + embodied-intelligence model.

Data: Mainland ChinaOverseas: None
Candidate (候选)Details

开悟 Kairos

上海大晓机器人Commercial

Cross-platform world-model base.

Data: Mainland ChinaOverseas: None
Candidate (候选)Details

World models

3 models

Galaxea G0

星海图Open weights / dataset

Whole-body-intelligence VLA for open scenarios.

Data: Mainland ChinaOverseas: None
Candidate (候选)Details

HY-Embodied

腾讯混元Research model

Tencent embodied-intelligence VLA model.

Data: Mainland ChinaOverseas: None
Candidate (候选)Details

影身 4D 世界模型

杭州影身智能Commercial

Full-stack 4D world model (deployed at a Vietnam shoe factory).

Data: Mainland ChinaOverseas: Yes (Vietnam factory)
Candidate (候选)Details

Diffusion & manipulation

3 models

GraspVLA / TrackVLA

银河通用Hardware-bound

Grasp and tracking VLA (bound to G1 platform).

Data: Mainland ChinaOverseas: None
Candidate (候选)Details

GR-Dexter

字节跳动 SeedResearch model

High-dexterity manipulation research model.

Data: Mainland ChinaOverseas: None
Candidate (候选)Details

RDT-1B

清华大学 TSAILMIT

Dual-arm diffusion manipulation research model.

Data: Self-hostableOverseas: None
Candidate (候选)Details

Hardware-bound commercial

1 models

傅利叶 GR 系列

傅利叶智能Commercial

Rehabilitation and general humanoid-robot model.

Data: Hardware-boundOverseas: Partial (overseas rehab)
Candidate (候选)Details

Models marked as “Candidate” are editorially shortlisted but not yet independently verified. Before bringing one into production, we recommend an independent compliance and deployment assessment.