By mid-2026, embodied intelligence remains one of the hottest sectors in the private markets. Even in a field this overheated, Da Xiao Robotics, which emerged only at the end of last year, has raised money at a rare pace. The company just announced that it has closed an angel+ round. Investors in the round included Dachen Capital, Shenzhen Capital Group, Shanghai Sci-Tech Innovation Fund, MuXi Co., Fosun RZ Capital, Hua Control Fund, and Lingang New Area Fund, among others. Existing shareholder SenseTime’s Guoxiang Capital also increased its stake, while Gejie Capital served as long-term financial adviser.

Including the angel round completed at the end of last year, Da Xiao has raised several hundred million dollars in a little more than six months. In an embodied-intelligence market that is getting hotter by the day, that capital is betting on a technical route that is also heating up fast: the world model. More and more people in the industry are coming to the same conclusion: if robots are ever going to operate in the real world, executing commands is nowhere near enough.

They also need to understand their environment, predict what changes next, and anticipate what may happen one step ahead. That is the job of the world model, the “brain” behind it all. Over the past six months, Da Xiao has rolled out a new answer for its Kairos world model roughly every few months: open-sourcing the model, enabling edge deployment, and launching a digital training arena. From model to deployment to training environment, Da Xiao’s integrated world-model approach, centered on understanding, generation, and prediction, is taking shape.

In recent days, the Kairos world model built on that architecture has ranked first in a series of embodied-intelligence benchmarks, including RoboTwin 2.0, LIBERO-Plus, WorldModelBench Robot, and DreamGen. In a sense, investors are not just betting on a story about world models. They are backing a technical path that is being continuously validated and increasingly delivered on.

Before this angel+ round, Da Xiao Robotics’ angel financing was led by Ant Group, with participation from Qiming Venture Partners, Kingkey Capital, Hony Capital, Lenovo Capital and Incubator Group, and Hanyuan Asset, the fund of Shanghai Jiao Tong University, among others. Existing shareholder SenseTime’s Guoxiang Capital continued to add to its position. Across the two rounds, Da Xiao brought together more than 15 investment institutions in six months. According to the company, the new funds will mainly go toward two directions: continuing to iterate on the world model and speeding up the rollout of key technologies such as large-scale simulation training and edge direct-drive control.

The second is building software-hardware integrated commercial solutions for smart retail, security inspections, cultural tourism, hotels, and other scenarios. Here is a quick primer on what a world model is. You can think of it as a large AI model that predicts what happens next. Like a large language model predicts the next word, a world model predicts the next frame and the physical changes inside it.

For robots, that amounts to a brain that can “fill in” the future. In the words of Da Xiao chairman Wang Xiaogang, embodied intelligence is the next era of artificial intelligence, and the “strongest brain” built around a world model is the key to opening that era. So how far has Da Xiao’s strongest brain come? Three releases in six months: from open source to edge deployment to a training arena, the company’s product rhythm has barely paused.

In December last year, as the company formally debuted, it also open-sourced Kairos World Model 3.0. It was the industry’s first open-source world model that supports commercial use, and it introduced for the first time an integrated “understanding-generation-prediction” architecture, bringing environmental understanding, future generation, and behavior prediction into one model framework. Meanwhile, Kairos 3.0 was also deeply adapted for domestic GPU makers including MuXi, Biren, and Sugon, connecting the full chain of “domestic model + domestic computing power.”

Three months later, Da Xiao launched Kairos 3.0-4B, a lightweight version with 4 billion parameters, and achieved the world’s first edge deployment of a world model. On NVIDIA’s Thor platform, which is designed for robotics, Kairos 3.0-4B reached a real-time generation speed of 1:1.5. For every second the robot experiences, it can predict 1.5 seconds into the future.

In a public demo, a robot running Kairos 3.0-4B completed a seven-minute household task in one uninterrupted take: tidying a coffee table, folding clothes, and preparing breakfast, all without human intervention. Behind that capability is Da Xiao’s continued work on model architecture, data systems, and inference efficiency. On the architecture side, Kairos keeps its integrated design, fusing understanding, generation, and prediction into the same network and directly mapping visual input to action output.

On the data side, Kairos 3.0-4B does not stop at conventional video training. In addition to video data, the model also takes in reasoning text that describes physical rules, human behavior demonstrations, and real robot operation trajectories. It learns not only how to do things, but why they should be done. On the efficiency side, the team used a self-developed attention operator to cut computational complexity from quadratic to linear.

Kairos is also compatible with NVIDIA as well as domestic GPU platforms from MuXi, Hygon, and Biren, and it supports different robot forms including single-arm, dual-arm, and dexterous-hand systems. But for embodied intelligence, models and data are still not enough. If robots are going to enter the real world, they also need a training environment large enough to prepare them for it. Against that backdrop, Da Xiao recently teamed up with the MMLab at the Chinese University of Hong Kong to launch Kairos-Homeworld, the world’s first unified framework for full-home 3D generation and object-level full interaction.

Kairos-Homeworld moves 300,000 real Chinese residential floor plans into the digital world, turning them into home environments where robots can train repeatedly. More importantly, Kairos-Homeworld is not just a massive household-scene dataset. It is also a data engine that keeps producing training environments. Through generative simulation, it can keep adding new scenes at very low marginal cost and continuously expand the scale of the environment.

Even more important, it brings Chinese homes into the robot-training system in a systematic way for the first time. The 300,000 real residential layouts are converted into training environments, and spatial structures common in Chinese homes, such as entryways, balconies, and enclosed kitchens, are also folded into large-scale training for the first time. With its built-in hierarchical generation framework, the system can automatically generate floor plans, arrange furniture, correct scenes, and assign physical properties.

Materials, density, friction coefficients, and other parameters are automatically modeled, allowing robots to train directly on grasping, carrying, and manipulation. From Kairos 3.0 to Kairos 3.0-4B to Kairos-Homeworld, Da Xiao has spent the past six months building out a full stack from model to deployment to training environment. That is the integrated technical loop capital is continuing to back.

A team that arrived just before the bell Da Xiao Robotics formally debuted only in early December last year. Chairman Wang Xiaogang is a co-founder of SenseTime and has long led the company’s technical work in computer vision and autonomous driving. Chief scientist Tao Dacheng is an academician of the Australian Academy of Science and ranks among the world’s most cited researchers in computer vision and machine learning. The team also brings together a group of frontier researchers from Nanyang Technological University, the University of Hong Kong, and the Chinese University of Hong Kong, with work spanning environment intelligence, world models, and embodied foundation models.

On the technical front, Da Xiao is building around a human-centered embodied-intelligence research paradigm and has created an end-to-end system that runs from environment-based data collection to the Kairos world model to a generalized embodied module. On the commercial side, the company has already launched its A1 embodied super-brain module, which can be adapted to mainstream platforms such as quadruped robots and supports end-to-end visual perception as well as natural-language instruction execution.

Its first target markets are smart retail and hotels. Based on the company’s roadmap, Da Xiao will next roll out the latest evolution of the Kairos world model, a large-scale data collection plan, and a full-stack hardware-software solution for retail and hotels.