Dexmal Launches Groundbreaking AI Model for Robotics

Dexmal's unveiling of its DM0 model is poised to redefine robotics as we know it. On February 10, the company introduced what it claims to be the world’s first embodied-native foundation model, a significant leap in how artificial intelligence can interact with the physical world.

At the Technology Open Day, Co-founder and CEO Tang Wenbin declared, “2026 is not the first year of embodied intelligence — it is the first year of ‘Embodied Native.’” This statement underscores a pivotal shift in AI development: moving beyond traditional models that simply interpret data to a framework that integrates AI deeply within real-world environments. This new paradigm emphasizes the importance of physical interaction in training and application, a concept that could have far-reaching implications for manufacturers, supply chains, and service industries globally.

Dexmal’s DM0 model has been meticulously designed with three core features: multi-source data pretraining, multi-task, cross-robot pretraining, and a spatial reasoning chain-of-thought mechanism. This sophisticated architecture allows the model to process diverse inputs and adapt its learning to a variety of robotic tasks. The company boasts that DM0 achieved remarkable results in the RoboChallenge, an authoritative robotics benchmark, where its 2.4 billion parameter version ranked first globally in both single-task and multi-task categories.

This accomplishment is not just about winning accolades; it reflects a fundamental shift in how robots can be utilized across industries. The traditional approach to AI in robotics often relies on pre-existing data sets that may not fully capture the nuances of real-world environments. Dexmal’s approach of starting from scratch allows for a more nuanced and adaptable model that learns directly from its interactions, potentially leading to robots that can operate more effectively in complex, dynamic settings.

For supply chain managers and executives, the implications are substantial. As robots become more adept at tasks ranging from assembly line work to logistics management, companies may find themselves reassessing how they implement automation. The flexibility and efficiency of embodied-native models like DM0 could reduce reliance on human labor for routine tasks, thereby altering workforce dynamics and operational costs. However, this also raises questions about the pace of adoption and the potential for job displacement.

Moreover, the emphasis on real-world applicability means that companies looking to integrate such technologies must consider the infrastructure needed to support them. This includes investments in training environments, data collection mechanisms, and ongoing maintenance. As Dexmal progresses, it will be critical for investors and policymakers to monitor how these embodied-native models affect the competitive landscape in manufacturing and beyond.

The cultural context surrounding robotics in China also plays a significant role in this development. The country has aggressively pursued AI and robotics as part of its broader strategy to become a global leader in technology. With government backing often involved, companies like Dexmal benefit from both funding and a favorable policy environment. This duality of state support and private innovation positions China uniquely in the global race for technological supremacy.

In conclusion, Dexmal's introduction of the DM0 model signals not just a technological advancement but a transformative moment in robotics. The shift towards embodied-native intelligence represents a new frontier that could redefine productivity, efficiency, and labor dynamics across numerous sectors. As companies seek to leverage these emerging capabilities, a careful analysis of their implications will be essential for navigating the complexities of the world's largest manufacturing ecosystem.

Sources

Dexmal Unveils DM0, the World’s First Embodied-Native Foundation Model