SenseTime unveils upgraded SenseFoundry platform, ushering in 'large model × vision AI' era

Chinese artificial intelligence company SenseTime unveiled a major upgrade to its flagship platform SenseFoundry at the 2025 SenseTime AI Forum, co-hosted with Hong Kong Science and Technology Parks Corporation on Dec. 9.

SenseFoundry is a one-top software platform tailored for Smart City management, addressing the needs of customers in the public sector. The company announced a comprehensive roadmap for bringing Vision AI 2.0 into large-scale deployment, signaling a new phase of transformation for smart cities and intelligent industry.

Today, Vision AI has become a central engine powering urban digitalization and industrial upgrades. However, traditional vision-AI development heavily depends on expert algorithm engineers and is hampered by long R&D cycles, high costs, and steep adoption barriers.

“The arrival of the large model era is accelerating a paradigm shift,” said Piao Yuankui, senior director of SenseTime’s Smart City and Enterprise Business Group. “Algorithm design no longer relies solely on specialists—now frontline engineers can participate directly. Applications are shifting from custom development to intelligent production.”

To tackle long-standing bottlenecks, SenseTime introduced the upgraded SenseFoundry Vision AI 2.0, which revolutionizes visual-algorithm production through two core systems:

1. General–Specialized Model Orchestration

SenseFoundry coordinates lightweight small models with general-purpose large models to perform progressive reasoning on long-tail visual tasks. This ensures high recognition accuracy while significantly reducing computing costs.

2. Intelligent Training System (Agentic Training)

Built around Agentic Training, the platform forms a full closed loop for data construction, model training, evaluation, and deployment. This enables end-to-end automation from image acquisition to business decision-making, allowing frontline engineers — not just experts — to rapidly build production-ready visual models.

Integrated vision-agent framework

SenseFoundry is also creating an integrated vision-agent framework that connects digital visual understanding with physical embodied action. This allows drones, quadruped robots, autonomous vehicles, and other heterogeneous edge devices to collaborate in perception and decision-making, enabling unified air–ground intelligence in inspection and patrol scenarios.

Powered by multimodal large models, SenseFoundry can efficiently schedule different types of vision models and run closed-loop workflows across scene perception, data processing, and intelligent decision-making. Its core advantage lies in pushing Vision AI beyond traditional boundaries — enabling systems that not only see, but also interpret, reason, and decide, supporting a shift in city governance from reactive response to proactive prediction.

As Vision AI 2.0 accelerates, SenseFoundry has already been deployed across more than 10 major categories of scenarios, including urban safety, transportation, manufacturing, drone patrol, and embodied intelligence, serving nearly 200 cities worldwide.

Feng Yu, general manager of SenseTime’s Hong Kong and Macao business, said Hong Kong’s ongoing Smart City Blueprint requires more systematic and integrated Vision AI capabilities. SenseFoundry’s upgraded platform — with its platformization, model-driven architecture, and agent-based design — aligns well with Hong Kong and Macao’s demand to move “from analysis to insight, and from insight to decision.”

Since entering the Asia-Pacific market in 2019, SenseTime has served nearly 500 international customers. With “Chinese-style innovation” gaining rapid global recognition, international visits to SenseTime’s Shanghai base surged fivefold this year alone. The combination of mature domestic experience and strong overseas localization capabilities is driving SenseTime’s Vision AI business into broader global markets.

From industry leader to vision AI trailblazer

With nine consecutive years as China’s market-share leader and industry recognition for its large-model vision capabilities, SenseTime is advancing AI from “usable” to “truly useful,” and from isolated point solutions to system-level intelligence.

Its upgraded SenseFoundry platform marks a milestone in the scaled deployment of vision large models and continues to push AI innovation into more industries and global regions.