Beijing is hosting its largest-ever auto exhibition, spanning 380,000 square meters with 1,451 vehicles on display and attracting millions of visitors. The 19th Beijing International Automotive Exhibition (referred to as the 2026 Beijing Auto Show) officially opened on April 24, bringing together nearly a thousand automakers from 21 countries and regions. A total of 181 new models made their global debut, while 71 concept cars were showcased, with industry leaders in full attendance.

Compared to scattered experiences at individual dealerships, ordinary consumers can more clearly perceive at this star-studded event how intelligence has penetrated every segment of the automotive industry. Family cars, rugged off-road vehicles, and commercial light trucks are all competing in autonomous driving capabilities. Models in the 300,000-yuan price range are competing on the number of laser radar lines, while those in the 100,000-yuan category are focusing on the quantity of laser radars. Automakers are emphasizing in-house research and development, while suppliers are striving to reduce costs and improve technical precision. New vehicles must have stable chassis, comfortable seats, front storage compartments, and spacious trunks.

As intelligence transitions from a premium feature to a standard requirement for all automakers, the next phase of competition in the automotive industry raises the question: what exactly are they competing on? Some answers may be found at the auto show.

Intelligent hardware is becoming more integrated. Currently, competition in the automotive industry is expanding from visible software and hardware configurations to more concealed underlying technical architectures. Automakers share a common goal: to save more costs and sell more vehicles.

"Integrated Cockpit and Driving" is one technical direction emerging from this competition. It involves combining the previously separate hardware, chips, and system architectures for intelligent driving and smart cockpit into a unified vehicle intelligence platform. This platform shares core computing power, underlying software, and electronic architecture, thereby achieving computing resource sharing and cost efficiency.

Around the time of the auto show, several companies announced new progress in integrated cockpit and driving technology. Intelligent automotive chip and solution provider Horizon recently launched its "Starry" series of vehicle intelligence chips and the "KakaClaw" onboard intelligence OS, using a single chip to manage both cockpit and driving functions, positioning it as a competitor to Tesla's "FSD+Grok" combination.

The Starry chip utilizes a castle security architecture to achieve hardware-level physical isolation between cockpit and driving functions. This not only meets the highest automotive safety standards, ensuring driving assistance systems remain unaffected by cockpit operations, but also enables dynamic computing resource reuse, allowing one set of hardware to simultaneously support driving assistance, voice interaction, entertainment navigation, and other full-scenario functions.

Horizon revealed that this solution can reduce vehicle components and space usage by nearly half, significantly lower memory requirements, and decrease hardware costs per vehicle by 1,500 to 4,000 yuan, accelerating the adoption of advanced intelligent configurations in mainstream 100,000 to 200,000 yuan models.

Intelligent automotive computing chip supplier Black Sesame Technologies showcased its mass-produced integrated cockpit and driving product, the "Wudang C1200 family," at the auto show. A single chip supports both smart cockpit and intelligent driving core functions. The displayed C1296 integrated domain controller and mass production solution "Dongfeng Tianyuan Smart Cabin Plus Platform" fully support large models, voice interaction, L2+ level assisted driving, and fully automatic parking assistance, with plans for installation in multiple production models.

"The electronic and electrical architecture in the automotive industry, like other electronics sectors, will inevitably evolve towards higher integration. The advent of central computing is certain, though the timing for technological convergence and implementation is still being explored," stated Black Sesame Technologies' CMO Yang Yuxin in an interview. He noted that intelligent configurations are continuously moving downmarket, with features previously found in 150,000-yuan models now appearing in 120,000, 100,000, and even 70,000-yuan vehicles, predicting that sub-70,000 yuan models will gradually standardize these features.

"The core of intelligence lies in the smart cockpit and intelligent driving. To help OEMs achieve standardization while controlling costs, architectural innovation is essential. Simply squeezing the supply chain for cost reduction has limits; architectural optimization is needed for win-win outcomes," Yang added.

Black Sesame's approach involves breaking through "from the bottom up," first implementing integrated cockpit and driving solutions in entry-level models. This replaces two or three systems with a single box or system, making intelligent configurations accessible for 100,000-yuan class vehicles.

In contrast, the current industry standard maintains clear technological divisions across price segments. Models above 120,000 yuan typically use a single-box, dual-chip solution (housing both driving and cockpit chips in one box), allocating capabilities based on product positioning without true chip integration. Models above 200,000 yuan more commonly employ a dual-box, dual-chip approach.

Yang Yuxin believes that with technological iteration, the single-box model will gradually move upmarket. "But integrated cockpit and driving must start with entry-level models, first meeting the demand for undifferentiated, cost-effective intelligent standardization."

Large models need to better understand the real world. In the AI era, another chain effect of technological democratization is the shift in consumer focus from "whether assisted driving is equipped" to "how capable it is." Consequently, automakers are now competing on refining model self-learning capabilities and scene replication accuracy.

"The challenge with world models lies in simulation fidelity and precision. If virtual scenes don't match the real world, training becomes meaningless; if model precision is insufficient, vehicle capability improvement will be distorted," said Peng Jun, founder and CEO of leading autonomous driving firm Pony.ai.

The "precision" Peng referred to denotes the closeness of the virtual environment, built on world models, to the real world. Higher precision results in more realistic virtual driving environments and vehicle performance that better mirrors reality.

For Pony.ai, which has focused on world models for five to six years, this technology is not a gimmick but a key safeguard for L4 autonomous driving safety. Real-world road testing struggles to cover vast numbers of extreme and edge traffic scenarios, while world models can fill these gaps, more comprehensively validating vehicle performance in complex interactive situations.

Leveraging this capability, Pony.ai's L4 autonomous driving solution has entered the urban delivery sector. During the auto show, the company jointly with CATL unveiled the world's first fully automotive-grade, fully redundant L4 unmanned light truck. It directly utilizes Pony.ai's urban road capabilities and operational system from its Robotaxi service, offering a range of 320-450 km and a top speed of 70-100 km/h.

Another company focused on autonomous driving solutions, Momenta, has率先 implemented physical AI in the L2 assisted driving sector. Momenta CEO Cao Xudong explained that physical AI is an autonomous driving technology paradigm based on world models and reinforcement learning. The core goal is to upgrade autonomous systems from "imitating human driving" to "understanding the physical world and evolving autonomously," ultimately achieving scalable data and commercial closed loops.

He illustrated that if large language models rely on Next Token Prediction to compress common sense of the digital world, giving AI text and natural language understanding abilities, world models use World Model Prediction to forecast future states and interaction logic of the physical world, enabling understanding of objects' physical properties, causal relationships of motion, and potential interactions.

Based on this logic, Momenta's R7 reinforcement learning world model can actively explore, repeatedly trial-and-error, and self-optimize in highly realistic virtual environments. Joint venture models like the SAIC Volkswagen ID. ERA 9X, SAIC Audi E7X, and Beijing Hyundai "IONIQ Venus" have announced they will feature Momenta R7-powered assisted driving systems.

Given the current competitive landscape, Cao Xudong predicts that only 2-3 suppliers in China and 3-4 globally will quickly emerge victorious in the intelligent driving field, with the industry structure rapidly converging.

AI needs more "emotion." As artificial intelligence technology evolves, more practitioners are considering and practicing how AI can become more human-like across various dimensions.

This month, smart vehicle full-stack AI solution provider SenseTime's Auto Intelligence division announced the launch of a new home AI smart terminal called "Care U." Technically, Care U is equipped with SenseTime's self-developed spatial multimodal interaction system, capable of recognizing faces, gestures, and the environment, while comprehensively capturing spatial motion states.

Through an end-cloud collaborative architecture, Care U possesses both rapid response capabilities and deep thinking abilities. Paired with a 1.75-inch OLED screen and 360-degree dynamic rotation mode, it can display over 50 dynamic expressions, making interactions more vivid.

Horizon's aforementioned large language model assistant "KakaClaw" is also described as a "more human-understanding" operating system. Reportedly China's first vehicle intelligence body OS, it implements a "task-as-a-service" smart interaction paradigm, supporting natural language commands to parallelly schedule driving and cockpit functions, automatically planning and executing cross-domain operation processes. It is characterized as "having personality, good memory, and multi-skilled."

At the launch event, Horizon founder and CEO Yu Kai specifically explained that "having personality" means KakaClaw supports multi-dimensional personality settings, offering switchable personas like the Doer, the Confidant, and the All-rounder, with support for dialects and full-scenario tone synchronization. "Good memory" refers to a long-sequence memory system that continuously learns user habits, automatically generating skills for personalized scenarios. "Multi-skilled" is evident in its ability not only to perform tasks and entertain but also to self-improve. Users can create专属 skills with zero code; both official and user-created skills can be published and shared, with the system automatically activating the most suitable skill combination.

In simple terms, beyond manual preference settings, systems that autonomously perceive and adapt to user habits in daily scenarios represent a key exploration direction for intelligent technology upgrades.

Matching hardware carriers are also continuously evolving, transitioning from wearables and small home terminals to more comprehensive, highly integrated large intelligent products.

The new Li Auto L9 Livis flagship SUV has thus become one of the most anticipated models at the 2026 Beijing Auto Show. Livis, originally an AI glasses product launched by Li Auto in December 2025, is seen as a key entry point for the company to integrate interactions inside and outside the cockpit.

Although Li Auto has not yet disclosed specific intelligent configurations for the new vehicle, the act of transplanting the embodied intelligence hardware Livis concept onto an entire vehicle indicates its ambition to upgrade its products from mere carriers to embodied intelligent terminals capable of perceiving, understanding, and deeply interacting with the physical world.

The automotive industry has inevitably merged with the AI wave. From optional features to standard equipment, from passive settings to active understanding, more human-centric intelligent experiences are redefining the future of mobility.