Gasgoo Munich- On November 5, Xie Tian, Baidu vice president and general manager of Baidu Maps, told the 2026 World Internet Conference Wuzhen Summit that Baidu Maps is forging a dual identity. It serves travelers directly as a spatiotemporal intelligence application — and feeds spatial perception, temporal inference, and smart decision-making capabilities into industries as a spatiotemporal intelligence base, underpinning applications in automotive, transportation, logistics, and tourism.
Drawing on two decades of data and technology, Baidu Maps has amassed over 13 million kilometers of true lane-level road data and more than 300 million points of interest. It logs over 600 million user interactions monthly, ships in over 60 million vehicles, and handles up to 650 billion Beidou high-precision positioning requests daily.
This data reservoir and at-scale service practice provide the foundation to continuously understand the real world — linking roads, locations, traffic conditions, and travel demand.
On top of that, Baidu Maps has built a spatiotemporal intelligence base composed of a data layer, a model layer, and an agent layer. The data layer continuously refreshes real-world information; the model layer conducts unified spatiotemporal modeling via MoT, with STI-core supporting inference and planning, and STI-diffusion driving dynamic evolution; the agent layer translates these capabilities into user services through products like the AI Copilot and Xiaodu Xiangxiang.
Through continuous data updates and model iterations, the base is constantly sharpening its grasp of real-world environments — and its ability to serve them.
In spatial services, Baidu Maps now covers over 200 countries and regions overseas. It has partnered with major automakers to map more than 1 million parking lots, enabling seamless spot-to-spot navigation.
For road risk perception, Baidu Maps detects hard braking and abnormal parking to warn trailing vehicles. In May, after a collision and fire at a tunnel exit in a southern Chinese city, the system issued continuous hazard alerts — logging roughly 13,000 warnings that day.
The platform also applies dynamic perception and forecasting to crowd flows at scenic spots and parking availability, giving travelers real-time reference points.
On information verification, Map AI taps transit, ticketing, and operating-hour data to answer specific queries — such as where to board a night cruise in Wuzhen's West Scenic Zone and the latest departure time, or how to time a cable car to catch the sunrise at Huangshan. The result: traceable, verifiable travel guidance.
As large models and agents push deeper into business workflows, the callers of map capabilities are shifting from humans to AI. Tasks once performed manually — finding locations, plotting routes — must now be translated into capabilities that agents can understand, invoke, and compose.
That means developers can embed map capabilities directly into AI applications, allowing agents to factor in real locations, traffic conditions, and scheduling when executing tasks.
To meet this shift, Baidu Maps is upgrading integration across three layers: ecosystem, product, and tool. At the ecosystem layer, it is driving integration with major AI platforms for seamless agent calls to map services. At the product layer, it offers native agent support and zero-code skill packages to lower the barrier to entry. At the tool layer, it launched MAPYA — a natural-language code generator — and the Baidu Maps CLI, a direct channel for AI programming tools, vastly simplifying development.
In these integration scenarios, developers can connect with just two commands, slashing development cycles from two days to mere minutes.
Baidu Maps now hosts over 4 million developers, working to turn map capabilities into off-the-shelf components that AI applications can call directly.
