Skip to main content

Dexmal

Category: Robotics / Embodied AI

An embodied AI startup developing end-to-end intelligent software and hardware systems to give robots human-like dexterity and reasoning through vision-language-action (VLA) models. Dexmal was founded in 2024. The company is led by Tang Wenbin. Based in Beijing, China. Team size: 50-100. Total funding raised: $140.0M. Latest round: Series A+, $55.5M, 2024-11, led by Alibaba Group. Key investors include Alibaba Group, NIO Capital, Legend Capital, Qiming Venture Partners, Lenovo Capital, Hongtai Fund, Jiukun Venture Capital.

Founded
2024
Headquarters
Beijing, China
Team size
50-100
Total funding
$140.0M

Value proposition

Lowers the barrier to general-purpose robotics by providing a unified VLA toolbox and open-source hardware, enabling robots to understand complex verbal/visual commands and execute long-horizon tasks.

Products and solutions

Dexbotic: An open-source Vision-Language-Action (VLA) software toolbox built on PyTorch., MemoryVLA: An AI model designed to solve temporal dependency challenges in long-duration robotic tasks., DOS-W1 (Dexbotic Open Source - W1): A modular open-source robot hardware platform., MMLA (Multi-Modal Large Model): A proprietary foundation model fusing multi-sensor inputs with LLM and world-model capabilities.

Unique value

Led by Megvii's former CTO and 'Yao Class' alumni, the company employs a rare 'Algorithm + Hardware + Application' DNA to solve the 'last-mile' problem of physical intelligence.

Target customer

Robotics researchers, developers, industrial automation firms, and hardware manufacturers requiring intelligent control systems.

Industries served

Robotics & Automation, Artificial Intelligence Research, Industrial Manufacturing, Consumer Electronics (Service Robots)

Technology advantage

Maintains a competitive lead by open-sourcing its core VLA infrastructure (Dexbotic) to establish industry standards while utilizing proprietary 'MemoryVLA' models to outperform competitors in complex, time-sensitive tasks.

How they differentiate

Unique 'open-source-first' strategy providing both a VLA software toolbox (Dexbotic) and modular hardware (DOS-W1), combined with proprietary 'MemoryVLA' models that solve temporal dependency in long-horizon tasks.

Main competitors

Physical Intelligence, Figure AI, NVIDIA (GROOT/N1), Google DeepMind (Gemini Robotics)

Key partnerships

Alibaba Group (Lead Investor/Strategic Partner), NIO Capital (Lead Series A Investor), RoboChallenge Committee (Co-founder for industry standardization), Legend Capital, Qiming Venture Partners, and Lenovo Capital (Strategic Financial Partners)

Notable customers

Alibaba Group (Logistics/Cloud Robotics), NIO (Factory Automation/Humanoid Testing), Global Robotics Research Labs (using Dexbotic/DOS-W1)

Major milestones

Founded in early 2024 by former Megvii CTO Tang Wenbin, Raised nearly 1 Billion Yuan ($140M) within its first year of operation, Released 'Dexbotic', a comprehensive open-source Vision-Language-Action toolbox, Launched 'MemoryVLA', a state-of-the-art model for complex, time-dependent robotic tasks, Co-launched 'RoboChallenge' evaluation platform with Hugging Face for industry standardization

Growth metrics

Scaled from founding to a team of ~100 employees within one year; Dexbotic open-source toolbox has become a standard for VLA researchers.

Market positioning

Infrastructure and foundation model leader in the Embodied AI sector, shifting focus from digital vision to physical robotics intelligence.

Geographic focus

Headquartered in China (Beijing/Shenzhen) with a global reach through open-source software and researcher communities.

Patents and IP

Specific patent numbers are not publicly disclosed, but the company actively publishes technical reports on ArXiv regarding its 'RoboChallenge' and VLA frameworks.

About Tang Wenbin

Tang Wenbin was the Co-founder and Chief Technology Officer (CTO) of Megvii Technology (Face++), one of China's 'Four Little Dragons' of AI, for over 13 years. He is a graduate of Tsinghua University's elite 'Yao Class' and is a world-renowned competitive programmer, having won a gold medal at the International Olympiad in Informatics (IOI) and achieved a 6th-place finish in the ACM-ICPC World Finals. He is widely recognized as one of the most influential technical leaders in China's AI and robotics sectors.

Official website: