Artificial intelligence is undergoing a fundamental shift as leading foundation models move beyond language processing toward understanding and predicting the physical world, the Beijing Academy of Artificial Intelligence (BAAI) said in its annual technology outlook released on Wednesday.
In its report, 2026 Top Ten AI Technology Trends, BAAI said the focus of AI development is moving away from expanding model size toward systems that can model causality, space and time — a change it described as a new technological paradigm.
“The competition is no longer about how large the parameters are,” BAAI Director Wang Zhongyuan said. “It is about whether models can understand how the world works. We are moving from predicting the next word to predicting the next state of the world.”
The report said this shift is being driven by the emergence of “Next-State Prediction” models, which aim to learn how environments evolve over time, enabling machines to reason, plan and act in physical settings rather than simply recognize patterns in data.
BAAI said 2026 would mark a turning point for the industry as AI moves from digital environments into physical ones and from research demonstrations into commercial deployment.
One major trend identified in the report is the rise of so-called world models that can learn physical laws and simulate real-world dynamics. These models are becoming important for applications such as autonomous driving, robotics training and industrial automation.
BAAI cited its own multimodal world model, WuJie, as an example of the approach.
The report also highlighted the rapid development of embodied intelligence, as AI systems are increasingly integrated into physical machines such as robots. It said humanoid robots are beginning to move from laboratory settings into controlled industrial environments.
At the same time, standardized communication protocols are allowing multiple AI agents to work together as coordinated systems, which the report said would be critical for handling complex tasks in scientific research and industrial production.
BAAI said the industry is also shifting its focus toward commercial value after an early wave of experimentation. On the consumer side, technology companies are building integrated AI platforms that combine assistants, services and applications into unified entry points. On the enterprise side, companies are demanding clearer returns on investment, pushing developers to focus on data quality, system integration and industry-specific products.
The report said artificial intelligence is also becoming more deeply integrated into scientific research, with the emergence of autonomous “AI scientists” that can generate hypotheses, design experiments and analyse results.
It said China should accelerate the development of independent scientific foundation models to reduce reliance on foreign technology and strengthen its position in emerging research fields such as materials science and drug discovery.
BAAI also pointed to the growing use of synthetic data as a response to the rising cost and limited availability of real-world training data, particularly in fields such as autonomous driving and robotics.
The report rejected the idea that AI innovation is slowing, saying inference optimisation and hardware efficiency remain key areas of progress, enabling more advanced models to run on edge devices and in resource-constrained environments.
It also highlighted the importance of open-source compiler ecosystems and software platforms compatible with different types of hardware to reduce supply chain risks and lower development barriers.
Finally, BAAI warned that AI security risks are becoming more complex, shifting from obvious errors toward more subtle forms of manipulation and deception, and said safety efforts would need to focus on model interpretability and adaptive defence systems.
“Security is becoming a core requirement for AI deployment,” the report said.
