As the 2026 Beijing Auto Show kicked off, competition in the automotive intelligence field has evolved from simple "voice assistants" to a new stage of "human-like intelligence." On the opening day of this exhibition, Volc Engine officially launched a new generation of automotive AI solutions based on Agentic AI architecture. This solution not only marks the implementation of the industry's first end-to-end AI cockpit architecture, but also announces that the number of large model installations has exceeded 7 million vehicles.

From "Chat Window" to "All-Round Butler": A Complete Architectural Revolution

Previously, the smart cockpits we were familiar with mostly adopted a "intent domain + multi-Agent collaboration" model, and users often felt obvious interaction boundaries and mechanical sensations. The solution launched by Volc Engine this time completely breaks through this limitation. Through a unified "AI brain," the new system deeply integrates key functional areas such as vehicle control, navigation, and intelligent driving.

image.png

This architectural upgrade brings a "perception, reasoning, execution, memory, learning" closed loop. Unlike traditional "question-and-answer" interactions, Agentic AI has autonomous driving capabilities. It can perceive environmental changes in real-time and independently advance processes based on task goals. For example, during long drives, the system can automatically switch to modes such as playing songs, telling stories, or showing animated movies based on the state of passengers in the back seat, without requiring the driver to repeatedly issue commands.

Three Engines Driving: Giving the Cockpit "Emotional Intelligence" and "Memory"

To make cold machines more like humans, the new solution incorporates three core engines:

  • Dialogue Reasoning Engine: Achieves always-on and full-duplex voice interaction. This means users can interrupt and join conversations anytime, just like chatting with friends, and the system can accurately identify and naturally participate in the dialogue.

  • Goal-Driven Engine: Grants the system the ability to handle complex scenarios. It can autonomously call on all vehicle tools based on environmental feedback, truly completing "cross-scenario tasks" from start to finish.

  • Learning and Growth Engine: The system is no longer "read and forgotten." It can accumulate experience during task execution. In addition to remembering user preferences, it can continuously self-summarize and evolve more efficient service skills.

Market Layout: Covering Over 50 Brands, Leading in Installation Volume

At the press conference, relevant officials revealed the latest industry achievements. Currently, more than 50 automotive brands and 145 models have been equipped with the Duanbao large model, with a total of over 7 million vehicles, and daily completion of cockpit interactions and service loops exceeding 30 million times.

During the auto show, several major new cars, including Mercedes-Benz electric GLC, SAIC Volkswagen ID. ERA9X, and the new "Jiayue" series from Rongwei, showcased their latest experiences after being equipped with this model.

Boundaries and Safety: Data Compliance and Cockpit-Driver Integration

Regarding privacy and driving safety, which have attracted public attention, relevant officials clearly stated that the automotive sector is a core strategic area, and data security remains the top priority. User data within the vehicle will never be used for training large models.

In the trend of "cockpit-driver integration," the role of the AI assistant is very clear: it is responsible for understanding intentions and issuing commands, but does not directly interfere with driving operations. The final driving loop is still completed by automakers and intelligent driving suppliers. This "intention understanding + data connectivity" collaboration model is expected to be implemented by the end of this year.