Developer docs
API playgroundTry for free, no card

Search company profiles

BeingBeyond

Full company profile

uuid00srme2

Namestring
BeingBeyond
Legal namestring
北京智在无界科技有限公司
Websiteurl
beingbeyond.com
Company typeenum
Private
Founded yearint
2025
Descriptiontext

BeingBeyond (智在无界) is a Beijing-based AI startup founded in May 2025 that develops foundation models for embodied intelligence, training general-purpose robot models on large-scale human video data rather than expensive robot demonstration data. The company's core technology stack comprises four foundation model series: Being-VL (cross-modal perception with 3D spatial understanding), Being-H (Vision-Language-Action model for dexterous hand manipulation pretrained on 150 million human hand samples via the UniHand dataset), Being-M (whole-body motion generation supported by the MotionLib million-scale dataset), and Being-W (low-level whole-body control via reinforcement learning). The platform also includes Being-H0.7, an implicit spatial world model trained on 200,000+ hours of human video, and supporting infrastructure such as MotionLib, UniHand, and the Universal Async Chunking (UAC) inference optimization enabling edge deployment.

The product portfolio spans two hardware products and two software solutions. D1 is a 19-DOF desktop-level dexterous robotic arm priced at the ten-thousand-yuan level targeting university and research buyers; U1 is the world's first Real DexUMI device integrating dexterous-hand data collection and robotic end-effector functionality with sub-millimeter tracking accuracy. Being-Actor is a whole-body teleoperation system for humanoid robots, and Being-Dex is a closed-loop framework enabling 30-minute new task learning on real robots. Revenue is generated through three streams: hardware sales (D1, U1) via direct sales engagement, enterprise model licensing for the Being-H-Flash world model product family (aura, ventus, procella tiers), and recurring model deployment, customization, and support services for robot manufacturers.

The company serves four primary customer segments: humanoid robot manufacturers seeking cross-embodiment generalizable foundation models, universities and research institutions procuring affordable research platforms, industrial manufacturing partners requiring precise dexterous manipulation, and industrial/service robot deployers needing cost-effective edge-deployable world models. Strategic partnerships include a chip-model co-optimization alliance with Digu Robotics (地瓜机器人) for edge deployment on Horizon S600/S100 chips, and academic collaborations with Peking University and Renmin University of China that have produced ICML 2025 and ICCV 2025 publications. The founder, Lu Zongqing (卢宗青), is a tenured associate professor at Peking University School of Computer Science and senior area chair at ICML/NeurIPS/ICLR.

Short descriptiontext

BeingBeyond (智在无界) is a Beijing-based AI startup founded in May 2025 that develops foundation models for embodied intelligence, training robot models on large-scale human video data and serving humanoid robot manufacturers, research institutions, and industrial deployers with models, hardware, and edge-deployable world model products.

Operating statusenum
Operating
Ownership categoryenum
Headcount rangeband
1–10
akta.pro rankint
HeadquartersBeijing Shi, China
HQ citystring
Beijing Shi
HQ countrystring
China
HQ regionstring
Asia
Markets served

Serves global market

Offices1 record

Each record includes

City, Country, Type, Description, Source

Keyword5 values
embodied intelligence models, humanoid robot foundation models, vision-language-action models, dexterous robotic manipulation, robot motion generation
Industry3 codes
PRIMARYHDAAAIAK
Robot Software, Autonomy Stacks & Simulation (ROS, digital twins)
SECONDARYHDAAAIAL
Robot Hardware, Actuators & Components (sensors, grippers, motors)
SECONDARYHDAAAIAJ
Humanoid & General-Purpose Robots
NAICS code2 codes
  • Computer Systems Design and Related Services54151
  • Computer and Peripheral Equipment Manufacturing33411
SIC code2 codes
  • Services-Prepackaged Software7372
  • Services-Computer Integrated Systems Design7373
Product category
Embodied Intelligence Foundation Models
Social media profiles1 record
GTM motion2 records

Each record includes

Type, Description, Source

Revenue model3 records
1Hardware Products (D1, U1)
TypeHardware Sales
Description

BeingBeyond generates revenue through sale of integrated hardware products including D1 (world's first desktop-level dexterous robotic arm priced at ten-thousand-yuan level) and U1 (Real DexUMI device). Contact sales at sales@beingbeyond.com.

beingbeyond.com
2Embodied Intelligence Brain Model Services
TypeSubscription Recurring
Description

Model deployment, function customization, technical support, and system upgrade services based on the company's proprietary embodied intelligence brain models for robot manufacturers and enterprise customers.

beingbeyond.com
3Model Licensing (Being-H-Flash)
TypeLicensing Royalties
Description

Licensing of world model products (Being-H-Flash product family: aura, ventus, procella) for deployment on edge AI chips, with tiered product offerings from general-purpose to deeply customized solutions for specific robot platforms and use cases.

beingbeyond.com
Marketing channels6 records

Each record includes

Title, Type, Stage, Description, Source

Distribution channels3 records

Each record includes

Title, Type, Scope, Target buyer, Description, Source

Cost components6 values
Personnel, Technology or R&D, Infrastructure, Marketing or Sales, Supply Chain, Operations
Pricing details3 tiers
1D1 Desktop Dexterous Robotic Arm — 10,000 yuan level
ModelUnit PricingBilling cadenceOne time/ perpetual license
Notes

World's first desktop-level dexterous robotic arm integrating arm, hand, and vision system, priced at approximately 10,000 yuan (万元级别), targeting universities and research institutions.

beingbeyond.com
2Being-H-Flash Product Family — Tiered by capability
ModelSubscriptionBilling cadenceAnnual
Notes

Being-H-Flash product family includes three tiers: aura (general scenario), ventus (accelerated edge), procella (deeply customized enterprise). Procella supports 2-4 week customization adaptation for new chip platforms. Dual NVIDIA and domestic chip adaptation.

beingbeyond.com
3Being-H-Flash operational cost advantage
ModelUsage-basedBilling cadenceMonthly
Notes

Being-H-Flash operational cost approximately 150 yuan/month per robot (for tasks like 2000 packages/day scanning), representing 98% cost reduction vs explicit world model alternatives at 8000+ yuan/month.

beingbeyond.com
GTM typeB2B
B2B
Offering typeSoftware
Software
Brand1 of 9 records shown
1Being-H
Description

Hand motion generation model series focused on dexterous manipulation, pretrained on massive human manipulation videos

beingbeyond.com
+8 more records
Core offering1 text field

BeingBeyond develops and commercializes embodied intelligence foundation models for humanoid robotics, including four core model families (Being-VL, Being-H, Being-M, Being-W) trained on large-scale human video data, plus integrated hardware products (D1 desktop robotic arm, U1 Real DexUMI) and software solutions (Being-Actor teleoperation, Being-Dex manipulation framework). The company sells these primarily as enterprise licenses and hardware units to robot manufacturers, industrial deployers, and research institutions.

Differentiator
Functional benefit
Problem solved
Quantifiable outcome1 of 7 values shown
  • Task success rate of 90%+ for new task learning on real robots in 30 minutes (Being-Dex)
+6 more records
Product overview1 text field

BeingBeyond offers a comprehensive embodied intelligence platform centered on four core foundation models (Being-VL, Being-H, Being-M, Being-W) trained on large-scale human video data. The product portfolio includes two hardware products (D1 desktop robotic arm and U1 Real DexUMI data collection system) and two software solutions (Being-Actor teleoperation system and Being-Dex manipulation solution). The foundation models form the AI brain: Being-VL handles cross-modal perception, Being-H generates dexterous hand motions, Being-M produces whole-body movements, and Being-W provides low-level motion control. Hardware products integrate with these models for data collection and deployment, while software solutions provide teleoperation and training frameworks for robot customers.

Product and service9 records
1Being-VL
CategoryFoundation Model / Software
Description

Next-generation cross-modal general-purpose foundation model for robotics. Based on multimodal perception of the physical world, it unifies 3D spatial understanding and human language intent, integrating perception, cognition, reasoning and planning for autonomous navigation, language interaction, and reliable task execution. Targeted at robot manufacturers and enterprise developers.

2Being-H
CategoryFoundation Model / Software
Description

Hand motion generation model series focused on dexterous manipulation. Pretrained on massive human manipulation videos to learn everyday human hand actions, enabling robotic dexterous hands to master new tasks with minimal real robot data. Covers fine assembly, flexible object arrangement, tool use, and complex interactions. Targeted at robot manufacturers and industrial deployers.

3Being-M
CategoryFoundation Model / Software
Description

Whole-body motion generation model trained on large-scale human videos. Serves as the bridge between robot brain and body, generating physically plausible, precise, and coherent full-body motion sequences including hand actions based on high-level task instructions. Generalizes across embodiments. Targeted at robot manufacturers building humanoid robots.

4Being-W
CategoryFoundation Model / Software
Description

Low-level whole-body control model designed for humanoid robots' complex dynamics. Trained with reinforcement learning on massive human data, it faithfully follows upstream full-body motion commands with stable zero-shot execution, addressing the core challenge of controlling humanoid robots. Targeted at robot manufacturers and humanoid robot developers.

5Being-Actor
CategorySolution / Software
Description

Whole-body teleoperation system for humanoid robots with 0.5B lightweight model covering 90% of daily actions. Supports voice, video imitation, and teleoperation control with zero code requirements. Applied to hazardous task operations and robot data factory for training. Targeted at robot manufacturers and researchers.

6Being-Dex
CategorySolution / Software
Description

Robot general dexterous manipulation solution with closed-loop system for data collection, model training, and deployment. Enables the base model to learn new tasks on a real robot within 30 minutes with 90%+ success rate. Achieves cross-embodiment generalization across different dexterous hands. Targeted at robot manufacturers and industrial deployers.

7D1 Desktop Dexterous Robotic Arm
CategoryHardware Product
Description

World's first desktop-level dexterous-hand robotic arm priced at ten-thousand-yuan level. Integrates 6-axis arm, 11-DoF dexterous hand, and 2-DoF active vision system (19 total degrees of freedom). Built with modular architecture, full-stack open-source Python API, and pre-equipped with Being-H VLA models covering the complete chain from data collection to deployment. Targeted at universities, research institutions, robot developers, and integrators.

8U1 Real DexUMI
CategoryHardware Product
Description

World's first Real DexUMI (Universal Manipulation Interface) for dexterous-hand data acquisition. Features palm-based isomorphic teleoperation glove, 11-joint design (6 active, 5 passive), 680g weight, 5kg payload, dynamic tracking error of <1mm with average of 0.1mm, and tactile sensors on five fingertips for multimodal embodied AI training. Targeted at robot manufacturers and researchers collecting dexterous manipulation data.

9Being-H-Flash World Model Product Family
CategoryLicensed Product / Software
Description

First world model product family designed with inference efficiency as a core capability. Includes three tiers: Being-H-aura (general scenario), Being-H-ventus (accelerated edge), and Being-H-procella (customized deep optimization for specific robot platforms and use cases). Achieves ~20 FPS real-time inference on 100+ TOPS edge platforms with dual adaptation for NVIDIA and domestic Chinese chips. Targeted at robot manufacturers and industrial deployers.

Scale indicator10 records

Each record includes

Type, Value, Description, Source

Partnership3 partners
1地瓜机器人 (Digu Robotics)
Strategic tierCoreTypeTechnology or IntegrationAnnounced on2026-06-11
Description

BeingBeyond and 地瓜机器人 (Digu Robotics) announced a strategic partnership on June 11, 2026, achieving the world's first world model real-time edge deployment. Being-H-Flash was successfully deployed on Digu Robotics' Sunrise S600/S100 edge AI computing chips, reaching ~20 FPS real-time inference on 100+ TOPS platforms. This partnership validates the 'chip-model co-optimization' approach and demonstrates that world models can run on edge chips rather than requiring data-center GPUs. Digu Robotics has over 500 SMB innovators, 500+ universities, and 100,000+ developers from 20+ countries on its platform.

beingbeyond.com
Strategic tierCoreTypeStrategic or Co-development Partner
Description

Joint research collaboration with Peking University to develop Being-VL-0.5, the first multimodal large model introducing Byte Pair Encoding into visual modality. Paper accepted at ICCV 2025. Collaboration also includes UniTacHand research (human-robot tactile data transfer) and Being-H0 physical instruction fine-tuning framework development.

Strategic tierCoreTypeStrategic or Co-development Partner
Description

Joint research with Renmin University of China on Being-M0, the first general motion generation framework with data-model collaborative scaling law characteristics. Paper accepted at ICML 2025. Collaboration established the complete 'data-model-migration' technical system and MotionLib million-scale dataset.

Recent move6 records

Each record includes

Date, Type, Title, Description, Source

Expansion highlight6 records

Each record includes

Type, Description

Peers10 records
TypeDirect peer
Description

US-based startup building foundation models for robot manipulation (π0). Most direct competitor to BeingBeyond's Being-H series, pursuing the same VLA-from-human-data paradigm for general-purpose robot dexterity.

TypeBroad incumbent
Description

Humanoid robot startup Figure AI develops its own Helix foundation model for general-purpose humanoid control. Comparable in pursuing general-purpose humanoid autonomy through foundation models with vertical integration.

TypeEmerging player
Description

Foundation models for warehouse robotics (acquired by Amazon). Adjacent comparable in applying foundation models to industrial robot deployment, though focused on warehouse picking rather than humanoid general manipulation.

TypeRegional player
Description

Leading Chinese humanoid and quadruped robot manufacturer. Direct customer-segment overlap as a Chinese humanoid OEM that would license BeingBeyond's foundation models for deployment on its robot platforms.

TypeBroad incumbent
Description

DeepMind's Gemini Robotics and RT-X initiatives pursue VLA foundation models trained on cross-embodiment robot data with the same goal of general manipulation. Most established incumbent in foundation models for robotics.

TypeDirect peer
Description

US startup building general-purpose robot foundation models (Skild Brain) targeting cross-embodiment generalization. Direct competitor pursuing the same embodied AI foundation model strategy from a US venture-backed position.

TypeRegional player
Description

Chinese embodied AI and humanoid robot company (Galaxea AI / Galbot) developing foundation models and humanoid platforms. Comparable in pursuing VLA-style foundation models with a Beijing-area technical base.

TypeBroad incumbent
Description

Tesla's Optimus humanoid robot uses internally developed foundation models trained on data from Tesla's autonomous driving fleet. Adjacent competitor developing proprietary embodied AI in-house rather than licensing third-party models.

TypeRegional player
Description

Chinese humanoid robot company pursuing embodied AI foundation models for its G1 humanoid platform. Most direct Chinese competitor with overlapping target customer base among domestic humanoid OEMs.

TypeBroad incumbent
Description

NVIDIA's Cosmos world foundation models and GR00T humanoid platform are the most cited incumbents in embodied AI. BeingBeyond explicitly benchmarks against Cosmos on cost (8000+ yuan/month vs 150 yuan/month) and trains on human video at scale similar to GR00T.

Market position
Strengths5 records

Each record includes

Headline, Details, Source

Weaknesses3 records

Each record includes

Headline, Details, Source

Competitive moat5 records

Each record includes

Type, Details

Key risks6 records

Each record includes

Headline, Details, Source

Key highlights7 records

Each record includes

Headline, Details, Source

Customer concentration

Classification, Details

Segment4 records

Each record includes

Title, Type, Primary, Description, Pain point addressed, Use case, Source

Ideal customer profile4 records

Each record includes

Profile, Firmographic size, Sales motion, Sales cycle length, Buying structure, Purchase trigger, Buyer persona, Geography, Industry vertical, Primary use case, Description, Pain points, Evidence proof points, Target buyer

Technology focused
Yes
API detail
Has APIbool
No

Docs URL, Description

AI capability10 records

Each record includes

Type, Description, Source

AI maturity
App detail

Has app

Feature11 records

Each record includes

Title, Differentiator, Description, Source

Core technology
Revenue estimate
Valuation estimate
Number of profiles
Profiles1 record

Each record includes

Name, Designation, Designation category, Overview, Profile commentary, Source

No data
No data
Funding overview

Funding stage, Last funding date, Total funding USD

Funding rounds

Each record includes

Round, Amount USD, Date, Pre money valuation, Total investors, Investors, News

Investors

Each record includes

Name, Type, Date of entry, Rounds participated, Website

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Each record includes

Name, Acquisition type, Announced date, Completed date, Status, Website, News

Investment

Each record includes

Name, Round, Announced date, Lead investor, Website, News

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

BeingBeyond

Embodied Intelligence Foundation Modelsbeingbeyond.com

BeingBeyond (智在无界) is a Beijing-based AI startup founded in May 2025 that develops foundation models for embodied intelligence, training robot models on large-scale human video data and serving humanoid robot manufacturers, research institutions, and industrial deployers with models, hardware, and edge-deployable world model products.

What BeingBeyond does

BeingBeyond (智在无界) is a Beijing-based AI startup founded in May 2025 that develops foundation models for embodied intelligence, training general-purpose robot models on large-scale human video data rather than expensive robot demonstration data. The company's core technology stack comprises four foundation model series: Being-VL (cross-modal perception with 3D spatial understanding), Being-H (Vision-Language-Action model for dexterous hand manipulation pretrained on 150 million human hand samples via the UniHand dataset), Being-M (whole-body motion generation supported by the MotionLib million-scale dataset), and Being-W (low-level whole-body control via reinforcement learning). The platform also includes Being-H0.7, an implicit spatial world model trained on 200,000+ hours of human video, and supporting infrastructure such as MotionLib, UniHand, and the Universal Async Chunking (UAC) inference optimization enabling edge deployment.

The product portfolio spans two hardware products and two software solutions. D1 is a 19-DOF desktop-level dexterous robotic arm priced at the ten-thousand-yuan level targeting university and research buyers; U1 is the world's first Real DexUMI device integrating dexterous-hand data collection and robotic end-effector functionality with sub-millimeter tracking accuracy. Being-Actor is a whole-body teleoperation system for humanoid robots, and Being-Dex is a closed-loop framework enabling 30-minute new task learning on real robots. Revenue is generated through three streams: hardware sales (D1, U1) via direct sales engagement, enterprise model licensing for the Being-H-Flash world model product family (aura, ventus, procella tiers), and recurring model deployment, customization, and support services for robot manufacturers.

The company serves four primary customer segments: humanoid robot manufacturers seeking cross-embodiment generalizable foundation models, universities and research institutions procuring affordable research platforms, industrial manufacturing partners requiring precise dexterous manipulation, and industrial/service robot deployers needing cost-effective edge-deployable world models. Strategic partnerships include a chip-model co-optimization alliance with Digu Robotics (地瓜机器人) for edge deployment on Horizon S600/S100 chips, and academic collaborations with Peking University and Renmin University of China that have produced ICML 2025 and ICCV 2025 publications. The founder, Lu Zongqing (卢宗青), is a tenured associate professor at Peking University School of Computer Science and senior area chair at ICML/NeurIPS/ICLR.

BeingBeyond firmographics

Firmographics
Name
BeingBeyond
Legal name
北京智在无界科技有限公司
Website
https://beingbeyond.com
Company type
Private
Founded year
2025
Operating status
Operating
Headcount range
1–10 employees
Short description
BeingBeyond (智在无界) is a Beijing-based AI startup founded in May 2025 that develops foundation models for embodied intelligence, training robot models on large-scale human video data and serving humanoid robot manufacturers, research institutions, and industrial deployers with models, hardware, and edge-deployable world model products.
Ownership category
akta.pro rank

BeingBeyond industry classification

Industry
Product category
Embodied Intelligence Foundation Models
NAICS
Computer Systems Design and Related Services (54151), Computer and Peripheral Equipment Manufacturing (33411)
SIC
Services-Prepackaged Software (7372), Services-Computer Integrated Systems Design (7373)
akta.pro industry
PRIMARYHDAAAIAK
Robot Software, Autonomy Stacks & Simulation (ROS, digital twins)
SECONDARYHDAAAIAL
Robot Hardware, Actuators & Components (sensors, grippers, motors)
SECONDARYHDAAAIAJ
Humanoid & General-Purpose Robots

Keywords

  • Embodied intelligence models
  • Humanoid robot foundation models
  • Vision-language-action models
  • Dexterous robotic manipulation
  • Robot motion generation

Where BeingBeyond is headquartered

Location

Headquarters

HQ city
Beijing Shi
HQ country
China
HQ region
Asia

Offices1 record

Markets served

BeingBeyond business model

Business model
GTM type
B2B
Offering type
Software
Cost components
Personnel, Technology or R&D, Infrastructure, Marketing or Sales, Supply Chain, Operations

Revenue model

  1. Hardware Products (D1, U1): BeingBeyond generates revenue through sale of integrated hardware products including D1 (world's first desktop-level dexterous robotic arm priced at ten-thousand-yuan level) and U1 (Real DexUMI device). Contact sales at sales@beingbeyond.com.
  2. Embodied Intelligence Brain Model Services: Model deployment, function customization, technical support, and system upgrade services based on the company's proprietary embodied intelligence brain models for robot manufacturers and enterprise customers.
  3. Model Licensing (Being-H-Flash): Licensing of world model products (Being-H-Flash product family: aura, ventus, procella) for deployment on edge AI chips, with tiered product offerings from general-purpose to deeply customized solutions for specific robot platforms and use cases.

Pricing tiers

ModelBillingPrice
Unit PricingOne time/ perpetual licenseD1 Desktop Dexterous Robotic Arm — 10,000 yuan level
SubscriptionAnnualBeing-H-Flash Product Family — Tiered by capability
Usage-basedMonthlyBeing-H-Flash operational cost advantage

Go-to-market motion2 records

Distribution channels3 records

Marketing channels6 records

BeingBeyond product offering

Product offering

Core offering

BeingBeyond develops and commercializes embodied intelligence foundation models for humanoid robotics, including four core model families (Being-VL, Being-H, Being-M, Being-W) trained on large-scale human video data, plus integrated hardware products (D1 desktop robotic arm, U1 Real DexUMI) and software solutions (Being-Actor teleoperation, Being-Dex manipulation framework). The company sells these primarily as enterprise licenses and hardware units to robot manufacturers, industrial deployers, and research institutions.

Product overview

BeingBeyond offers a comprehensive embodied intelligence platform centered on four core foundation models (Being-VL, Being-H, Being-M, Being-W) trained on large-scale human video data. The product portfolio includes two hardware products (D1 desktop robotic arm and U1 Real DexUMI data collection system) and two software solutions (Being-Actor teleoperation system and Being-Dex manipulation solution). The foundation models form the AI brain: Being-VL handles cross-modal perception, Being-H generates dexterous hand motions, Being-M produces whole-body movements, and Being-W provides low-level motion control. Hardware products integrate with these models for data collection and deployment, while software solutions provide teleoperation and training frameworks for robot customers.

Differentiator

Problem solved

Functional benefit

Brands

  • Being-H: Hand motion generation model series focused on dexterous manipulation, pretrained on massive human manipulation videos
  • Being-M
  • Being-VL
  • Being-W
  • Being-H-Flash
  • Being-Dex
  • Being-Actor
  • D1
  • U1

Products and services

  • Being-VL Next-generation cross-modal general-purpose foundation model for robotics. Based on multimodal perception of the physical world, it unifies 3D spatial understanding and human language intent, integrating perception, cognition, reasoning and planning for autonomous navigation, language interaction, and reliable task execution. Targeted at robot manufacturers and enterprise developers.
  • Being-H Hand motion generation model series focused on dexterous manipulation. Pretrained on massive human manipulation videos to learn everyday human hand actions, enabling robotic dexterous hands to master new tasks with minimal real robot data. Covers fine assembly, flexible object arrangement, tool use, and complex interactions. Targeted at robot manufacturers and industrial deployers.
  • Being-M Whole-body motion generation model trained on large-scale human videos. Serves as the bridge between robot brain and body, generating physically plausible, precise, and coherent full-body motion sequences including hand actions based on high-level task instructions. Generalizes across embodiments. Targeted at robot manufacturers building humanoid robots.
  • Being-W Low-level whole-body control model designed for humanoid robots' complex dynamics. Trained with reinforcement learning on massive human data, it faithfully follows upstream full-body motion commands with stable zero-shot execution, addressing the core challenge of controlling humanoid robots. Targeted at robot manufacturers and humanoid robot developers.
  • Being-Actor Whole-body teleoperation system for humanoid robots with 0.5B lightweight model covering 90% of daily actions. Supports voice, video imitation, and teleoperation control with zero code requirements. Applied to hazardous task operations and robot data factory for training. Targeted at robot manufacturers and researchers.
  • Being-Dex Robot general dexterous manipulation solution with closed-loop system for data collection, model training, and deployment. Enables the base model to learn new tasks on a real robot within 30 minutes with 90%+ success rate. Achieves cross-embodiment generalization across different dexterous hands. Targeted at robot manufacturers and industrial deployers.
  • D1 Desktop Dexterous Robotic Arm World's first desktop-level dexterous-hand robotic arm priced at ten-thousand-yuan level. Integrates 6-axis arm, 11-DoF dexterous hand, and 2-DoF active vision system (19 total degrees of freedom). Built with modular architecture, full-stack open-source Python API, and pre-equipped with Being-H VLA models covering the complete chain from data collection to deployment. Targeted at universities, research institutions, robot developers, and integrators.
  • U1 Real DexUMI World's first Real DexUMI (Universal Manipulation Interface) for dexterous-hand data acquisition. Features palm-based isomorphic teleoperation glove, 11-joint design (6 active, 5 passive), 680g weight, 5kg payload, dynamic tracking error of <1mm with average of 0.1mm, and tactile sensors on five fingertips for multimodal embodied AI training. Targeted at robot manufacturers and researchers collecting dexterous manipulation data.
  • Being-H-Flash World Model Product Family First world model product family designed with inference efficiency as a core capability. Includes three tiers: Being-H-aura (general scenario), Being-H-ventus (accelerated edge), and Being-H-procella (customized deep optimization for specific robot platforms and use cases). Achieves ~20 FPS real-time inference on 100+ TOPS edge platforms with dual adaptation for NVIDIA and domestic Chinese chips. Targeted at robot manufacturers and industrial deployers.

Quantifiable outcome

  • Task success rate of 90%+ for new task learning on real robots in 30 minutes (Being-Dex)
  • +6 more outcomes

Companies that use BeingBeyond

Customer profile

Segments4 records

Ideal customer profiles4 records

BeingBeyond technology and API

Technology

Technology focussed Yes

API detail

Has API
No
API docs
API detail

Core technology

AI maturity

App detail

AI capability10 records

Feature11 records

BeingBeyond partnerships and signals

Strategic signal

Partnerships

Three partnerships are on record, tiered core.

  • 地瓜机器人 (Digu Robotics)coreTechnology or Integration · 11 June 2026BeingBeyond and 地瓜机器人 (Digu Robotics) announced a strategic partnership on June 11, 2026, achieving the world's first world model real-time edge deployment. Being-H-Flash was successfully deployed on Digu Robotics' Sunrise S600/S100 edge AI computing chips, reaching ~20 FPS real-time inference on 100+ TOPS platforms. This partnership validates the 'chip-model co-optimization' approach and demonstrates that world models can run on edge chips rather than requiring data-center GPUs. Digu Robotics has over 500 SMB innovators, 500+ universities, and 100,000+ developers from 20+ countries on its platform.
  • 北京大学 (Peking University)coreStrategic or Co-development PartnerJoint research collaboration with Peking University to develop Being-VL-0.5, the first multimodal large model introducing Byte Pair Encoding into visual modality. Paper accepted at ICCV 2025. Collaboration also includes UniTacHand research (human-robot tactile data transfer) and Being-H0 physical instruction fine-tuning framework development.
  • 中国人民大学 (Renmin University of China)coreStrategic or Co-development PartnerJoint research with Renmin University of China on Being-M0, the first general motion generation framework with data-model collaborative scaling law characteristics. Paper accepted at ICML 2025. Collaboration established the complete 'data-model-migration' technical system and MotionLib million-scale dataset.

Scale indicators10 records

Recent moves6 records

Expansion highlights6 records

BeingBeyond competitors and assessment

Company assessment

Direct peers

  • Physical Intelligence: US-based startup building foundation models for robot manipulation (π0). Most direct competitor to BeingBeyond's Being-H series, pursuing the same VLA-from-human-data paradigm for general-purpose robot dexterity.
  • Skild AI: US startup building general-purpose robot foundation models (Skild Brain) targeting cross-embodiment generalization. Direct competitor pursuing the same embodied AI foundation model strategy from a US venture-backed position.

Broad incumbents

  • Figure AI: Humanoid robot startup Figure AI develops its own Helix foundation model for general-purpose humanoid control. Comparable in pursuing general-purpose humanoid autonomy through foundation models with vertical integration.
  • Google DeepMind (Gemini Robotics): DeepMind's Gemini Robotics and RT-X initiatives pursue VLA foundation models trained on cross-embodiment robot data with the same goal of general manipulation. Most established incumbent in foundation models for robotics.
  • Tesla (Optimus): Tesla's Optimus humanoid robot uses internally developed foundation models trained on data from Tesla's autonomous driving fleet. Adjacent competitor developing proprietary embodied AI in-house rather than licensing third-party models.
  • NVIDIA (Cosmos / GR00T): NVIDIA's Cosmos world foundation models and GR00T humanoid platform are the most cited incumbents in embodied AI. BeingBeyond explicitly benchmarks against Cosmos on cost (8000+ yuan/month vs 150 yuan/month) and trains on human video at scale similar to GR00T.

Emerging players

  • Covariant: Foundation models for warehouse robotics (acquired by Amazon). Adjacent comparable in applying foundation models to industrial robot deployment, though focused on warehouse picking rather than humanoid general manipulation.

Regional players

  • Unitree Robotics: Leading Chinese humanoid and quadruped robot manufacturer. Direct customer-segment overlap as a Chinese humanoid OEM that would license BeingBeyond's foundation models for deployment on its robot platforms.
  • Galbot: Chinese embodied AI and humanoid robot company (Galaxea AI / Galbot) developing foundation models and humanoid platforms. Comparable in pursuing VLA-style foundation models with a Beijing-area technical base.
  • AgiBot: Chinese humanoid robot company pursuing embodied AI foundation models for its G1 humanoid platform. Most direct Chinese competitor with overlapping target customer base among domestic humanoid OEMs.

Market position

Strengths5 records

Weaknesses3 records

Competitive moat5 records

Key risks6 records

Key highlights7 records

Customer concentration

BeingBeyond social profiles

Digital presence

BeingBeyond financial estimates

Financial estimate

Revenue estimate

Valuation estimate

BeingBeyond leadership team

Management profile

Number of profiles

Profiles1 record

BeingBeyond funding detail

Funding detail

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

BeingBeyond M&A and investment

M&A and investment

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about BeingBeyond

What does BeingBeyond do?

BeingBeyond develops and commercializes embodied intelligence foundation models for humanoid robotics, including four core model families (Being-VL, Being-H, Being-M, Being-W) trained on large-scale human video data, plus integrated hardware products (D1 desktop robotic arm, U1 Real DexUMI) and software solutions (Being-Actor teleoperation, Being-Dex manipulation framework). The company sells these primarily as enterprise licenses and hardware units to robot manufacturers, industrial deployers, and research institutions.

Is BeingBeyond a public or private company?

BeingBeyond is a private company. It is classified as unknown and is currently operating.

When was BeingBeyond founded?

BeingBeyond was founded in 2025. It employs 1 to 10 people.

Where is BeingBeyond based?

BeingBeyond is headquartered in Beijing Shi, China, in the Asia region.

How does BeingBeyond make money?

Three revenue lines are on record. Hardware Products (D1, U1) is the primary driver. The others are embodied Intelligence Brain Model Services and model Licensing (Being-H-Flash).

Who are BeingBeyond's main competitors?

Direct peers on record are Physical Intelligence and Skild AI. Broad incumbents are Figure AI, Google DeepMind (Gemini Robotics), Tesla (Optimus) and NVIDIA (Cosmos / GR00T). Covariant is listed as an emerging player. Regional players are Unitree Robotics, Galbot and AgiBot.

Does BeingBeyond have an API?

No public API is recorded for BeingBeyond.

What industry is BeingBeyond in?

BeingBeyond's product category is Embodied Intelligence Foundation Models. Its primary akta.pro industry code is HDAAAIAK, Robot Software, Autonomy Stacks & Simulation (ROS, digital twins), with a secondary code of HDAAAIAL, Robot Hardware, Actuators & Components (sensors, grippers, motors). Its NAICS code is 54151 and its SIC code is 7372.

Unlock the full company data

50 free credits on sign-up, no credit card required.

Contact sales
Live signals
PandailyBeingBeyond's Being-M0.7 Uses 10,000 Hours of Human Video to Teach Humanoids to Walk and GraspBeingBeyond released Being-M0.7, a latent world-action model pretrained on over 10,000 hours of human video and motion data. It achieved 71.1% on the SIMPLE simulation benchmark and tied the strongest baseline on a Unitree G1 robot with 86.7% success. The model predicts future visual latents and whole-body motion before acting.GasgooBeingBeyond and PaXini Form Comprehensive PartnershipBeingBeyond and PaXini signed a cooperation agreement, combining BeingBeyond's embodied foundation models with PaXini's tactile perception and data. BeingBeyond unveiled Being-H0.8, an implicit tactile world action model pre-trained on over 500,000 hours of human video, with PaXini supplying multimodal data. The partnership aims to define industry standards for embodied intelligence.GasgooSYNAPATH AI and BeingBeyond Deepen CooperationSYNAPATH AI and BeingBeyond signed a strategic partnership to deepen collaboration on data development, model training validation, and industry data standards. SYNAPATH AI has already contributed high-quality data to BeingBeyond's embodied foundation models, including the Being-H0.8 model. The companies will explore a data-sharing ecosystem platform for the sector.KuCoinBeing-M0.7: The World’s First Implicit World-Action Model for Full-Body Mobility in Humanoid RobotsBeingBeyond (Zhi Zai Wu Jie) has launched Being-M0.7, the world's first latent world-action model designed for full-body mobility and manipulation in humanoid robots, pre-trained on over 10,000 hours of human-centered multimodal data. The model successfully demonstrated challenging whole-body tasks including fluid interaction, mirror reasoning, long-range item retrieval, and obstacle avoidance, outperforming NVIDIA's GR00T-N1.6 in comparative tests. This release signals a shift in humanoid robotics competition from hardware capabilities to model-level scene understanding and coordinated full-body control, with implications for scaling embodied AI development.GasgooBeingBeyond and Linkerbot Establish Strategic PartnershipBeingBeyond, an embodied general foundation model developer, signed a strategic partnership with Linkerbot, a global leader in dexterous robotics, on July 13, 2026. The two companies plan to combine foundational AI models with dexterous hardware for dataset construction, pre-training development, and scenario deployment to scale embodied intelligence solutions. The partners will showcase their joint D1 desktop robotic arm at the World Artificial Intelligence Conference (WAIC) 2026, opening July 17.GasgooPNDbotics, BeingBeyond Form Strategic PartnershipPNDbotics recently signed a cooperation agreement with BeingBeyond to deeply integrate hardware and AI capabilities for humanoid robotics development. The two companies plan to collaborate long-term on dataset construction, pre-training model development, and robot application promotion, with the goal of accelerating large-scale deployment of humanoid robots across industrial scenarios. At the 2026 World Artificial Intelligence Conference, they plan to unveil Adam-U equipped with BeingBeyond's Being-H model, representing a phased result of their joint R&D efforts.GasgooD-Robotics, BeingBeyond Forge Strategic PartnershipD-Robotics and BeingBeyond announced a strategic partnership on June 11, completing the edge-side deployment of BeingBeyond's Being-H-Flash world model on D-Robotics's Ascend S600 high-performance computing chip. Real-world testing showed the optimized system achieves inference speeds approaching 20 FPS, enabling real-time AI processing directly on the device without cloud dependency. The collaboration sets an industry benchmark for real-time interaction in embodied intelligence applications.36krZhiZai WuJie and Peking University Team Achieve New Breakthrough: Endowing Robots with HumanBeingBeyond and Peking University researchers developed UniTacHand, a framework that transfers human tactile skills to dexterous robots with only 10 minutes of paired data. It achieved zero-shot lossless transfer, reaching 100% success in object positioning and 85% in classification tasks. The work addresses the data desert and morphological gap in robot tactile research.Pedaily一个GPT Plus会员的钱,够机器人跑一个月世界模型了Zhi Zai Wu Jie released Being-H-Flash, an implicit world model that runs in real-time on edge chips, costing 150 yuan monthly per robot. It is 2% of NVIDIA's Cosmos and 70% cheaper than Pi0.5, and is the first world model to achieve real-time operation on hundreds of TOPS edge chips.