At this year’s Guangzhou Auto Show, beyond the glitzy unveilings and concept cars, a quieter contest was unfolding inside dashboards: automakers and tech suppliers are racing to see whose large language model (LLM) becomes the operating system of tomorrow’s smart vehicles.
The signal is clear in 2025’s new mass-produced models. Doubao, the LLM from Volcano Engine, has emerged as the most widely deployed in the industry. From the all-electric Mercedes-Benz CLA and SAIC Audi E5 Sportback to Roewe M7 DMH, Dongfeng eπ 007, and Changan Mazda EZ60, Volcano Engine’s software was ubiquitous, serving as the busiest AI brain at the show.
This represents a pivot in the logic of smart vehicle competition. In earlier years, the race centered on raw computing power, sensor arrays, and chipsets. Today, it’s about whose LLM is more sophisticated, whose model better understands drivers and vehicles, and who can turn data into actionable intelligence. Volcano Engine and Doubao are rapidly becoming the industry’s poster child for this shift.
Two years ago, LLMs in vehicles were largely a marketing gimmick. Automakers would integrate an open API, produce a few demo videos, and claim to offer AI cockpits. Users, however, often encountered multi-second delays when requesting navigation or issuing commands. Many reverted to manually interacting with touchscreens.
Now, the narrative has changed. At the Guangzhou Auto Show, performance and utility mattered as much as presence. Doubao has been integrated across a broad spectrum of models, covering sedans, SUVs, hybrids, and EVs, with price points ranging from 100,000 to 300,000 yuan.
Industry power is shifting as well. The discussion has moved from chips and sensors to the strength of foundation models and the breadth of real-world usage data. LLMs are no longer a demo feature—they are becoming core infrastructure. Volcano Engine is positioning itself to evolve from a content platform into a Tier-1 AI supplier for automakers.
On the all-electric Mercedes-Benz CLA, Doubao powers a new intelligent human-machine interface. Commonly used features are auto-organized into cards on the dashboard homepage, improving interaction efficiency by 50%. Response times reach 0.2 seconds, with full task execution in 0.8 seconds. The system recognizes four distinct emotions, distinguishing playful comments from complaints—a sophistication previously unseen in luxury vehicles.
SAIC Audi E5 Sportback leverages Doubao for multi-device integration. The AI assistant interprets fuzzy commands and contextual requests. A phrase like “I’m about to visit a client, help me get ready” triggers scheduling, navigation, and vehicle adjustments seamlessly. The assistant supports multi-turn conversation, car control operations, and even voice cloning, creating a personalized digital companion.
Roewe M7 DMH demonstrates “deep reasoning” capabilities. Using the Doubao Deep Reasoning Model, it can interpret inverted sentences, double negatives, and nuanced intentions. It retains user preferences, frequented locations, and prior dialogues. During the launch, the system showcased a feature to “help put my child to sleep,” orchestrating temperature, fan speed, window closure, volume adjustments, and story playback—all through a single voice command.
Dongfeng eπ 007 and Changan Mazda EZ60 showcase more playful integrations. The eπ 007 allows up to ten high-frequency commands in a single sentence, triggering jokes, videos, or mini-games. It can convert dashcam footage into artistic images for social sharing. The EZ60 offers over 1,700 functions, connecting car controls, travel advice, and entertainment.
The takeaway is clear: LLM integration transforms vehicles from transport machines into AI assistants, streamlining complex tasks into natural-language commands. For automakers, this raises questions about control—who designs the user experience, the OEM or the supplier?
Volcano Engine has forged partnerships with most major automakers, including Mercedes-Benz, BMW, Volkswagen, SAIC, and GAC. Its influence extends beyond infotainment to intelligent driving, digital transformation, and backend services. For Mercedes-Benz, the collaboration now encompasses intelligent cockpits, big data, digital marketing, and full-lifecycle customer management. ByteDance’s ecosystem allows automakers to analyze data and optimize interactions, from pre-sale outreach to post-purchase retention.
The strategy represents a shift from content platform to industrial cloud and large model infrastructure. One end links to ByteDance’s content and user base, the other integrates automakers’ R&D, marketing, and service operations. Doubao acts as the bridge, with cars as nodes and computation shared between cloud and vehicle.
Yet the model presents challenges. Widespread deployment across multiple brands may limit differentiation. Automakers must consider data security, ownership, and whether they can switch providers or develop proprietary systems. Volcano Engine aspires to be the “Android plus cloud” of automotive AI—but OEMs may resist relinquishing control over their digital ecosystems.
At the Guangzhou Auto Show, Doubao claimed dominance in installations, but the competitive race is far from over. Success will depend on three factors: deeply integrating LLMs into specific vehicle use cases, rapidly iterating based on extensive in-vehicle data, and extending applications beyond the cockpit into higher-value services.
For suppliers like Volcano Engine, balancing scale with automaker autonomy is critical to avoid homogenization. For OEMs, the question becomes: what constitutes their core asset in an era of electrification and AI—battery, motor, or digital brain?
