Chery Unveils CheryMind AI Model and Ziwei Chips, Says It Is an AI Company

Yin, who has spent more than four decades in the auto industry, argued that the risk facing automakers is not that AI is too strong or too fast, but that they fail to adapt. He compared the shift to 1886, when Benz built the first car and the horse keepers of the era were left behind. 'The real risk is not that AI is too strong or too fast, but that we cannot continue to be the horse keepers of that era,' Yin said. He added that AI will eliminate companies that refuse it or are afraid to use it.

The centerpiece is CheryMind, a domain-specific foundation model that Chery says is built for factories and engineering rather than general conversation. According to Leiphone, it is trained on 700TB of enterprise knowledge covering technical standards, process specifications, design rules and safety norms; 200PB of engineering data from simulation, testing, CAD and vehicles; tens of millions of R&D documents; 25 million coaching-level driving data; and 5 million robot motion-capture data.

Chery said CheryMind is already in use across more than 1,500 end-to-end business processes, more than 400 connected enterprise systems, more than 60,000 daily employee uses and more than 200 business scenarios. Executive Vice President and CTO Gao Xinhua said an engineer previously had to run six rounds and spend more than 100 days on one simulation optimization; with CheryMind, each person can produce 100 plans in half a day, a 42-fold efficiency gain.

The model operates on two tracks. Enterprise vertical models cover CAD, quality, manufacturing and enterprise agents across R&D, manufacturing, quality, supply, marketing and operations. Product vertical models cover smart cockpits, smart driving, vehicle intelligence, flight control, MoAgent and an agricultural machinery operating system, extending to eVTOL, robots and farm equipment. Chery says new capabilities generated in real scenarios flow back to the base model, creating a closed loop.

Chery also unveiled its Ziwei chip family, named after the North Star. The C series handles computing. The C100 central computing chip has 2,400 TOPS of vehicle NPU computing power, 600GB/s memory bandwidth and support for 100B-parameter models deployed on the vehicle; it completes an end-to-end perception-to-decision algorithm loop in 80ms. Chery compared it with industry flagship products at 2.2 times the bandwidth, 1.2 times the computing power and 1.5 times the image processing performance.

The S series handles energy. The automotive-grade silicon carbide S1500 has a 1,500V withstand voltage, 1,450A output current, efficiency of at least 99.6%, a junction temperature of 200°C and more than 30% lower mass-production cost on 8-inch wafers. The M series handles control, with the MC200 domain controller, MS100 pre-driver, MX100 PMIC and MC100 electric drive scheduled to land from 2027. Chery said it has invested more than 110 billion yuan in R&D over five years, operates 313 Yaoguang laboratories and raised ADAS penetration by 267% in two years.

Beyond vehicles, Chery presented a 'one base × N scenarios' structure. Its CHERY AWE world engine uses three models. Goujing has built more than 700,000 scenarios, including more than 300,000 high-risk and 20,000 extreme scenarios, and encodes multi-view images, LiDAR point clouds, scene structure, description text and driving trajectories into a latent space for closed-loop simulation of long-tail cases. Guanyu handles multimodal reasoning with short-term memory and a language topology graph, allowing a vehicle to reach a destination using road signs and output a traceable reasoning chain. Xingce uses a MoT multi-expert architecture to shift from passive response to active risk prediction.

Falcon ADAS has run with this base for a year. Chery said it meets regulations in more than 130 countries and regions, has received C-ICAP five-star-plus and IVISTA G-plus certification, and won six consecutive China smart driving championships. Falcon 500 offers urban assistance as standard with the same intelligence for fuel and electric vehicles; Falcon 700 covers full-scenario park-to-park and cross-floor valet parking; Falcon 900 covers highway and urban L3 with redundancy in perception, control and power. The Lingxi 8D cockpit uses an AIOS architecture to connect chassis, cockpit, body, ADAS and powertrain domains, with millisecond cross-domain response, outside pedestrian-intent recognition and inside emotion and fatigue sensing. It also features 33 speakers, a 43.7-inch PHUD and S2S end-to-end voice.

Chery showed a native front-installed Robotaxi for global markets, with five native features covering architecture, front-installation, experience, operation and globalization, and six safety layers covering passengers, behavior, perception, algorithm, execution and fleet. It targets 100,000 vehicles in global operation by 2030. Robots reuse the company's 30-year supply chain, hardware architecture and quality system. Under a multi-agent architecture, MoAgent 2.0, based on CheryMind, uses a slow brain for understanding and planning and a fast brain for real-time response, with edge-cloud collaboration forming a data loop. Chery has laid out quadruped and wheeled robot dogs, police robots, medical guides, family companion robots and industrial and humanoid robots; its Moja robots have shipped more than 2,000 units to more than 60 countries.

In energy and agriculture, Chery said its Xunlong instant charging provides 500 km of range in five minutes, and a solid-state battery pilot line will be operational in the first half of 2027. Smart agricultural machinery cuts energy consumption by 74% and raises yield from 1,600 to 2,000 jin per mu, with a target of serving 1 billion mu of farmland globally by 2035. Chery is also opening its technology platform, scenarios and industrialization capabilities to suppliers, universities and startups, and is building a 2050 Future Academy with the University of Science and Technology of China for long-cycle, fault-tolerant frontier research.

Yin said technology only creates freedom when it leaves the laboratory and enters roads, homes, farmland and energy systems to solve real problems. 'Technology only creates freedom when it leaves the lab and enters roads, homes, farmland and energy systems to solve real problems,' he said.

Editor's Summary

Chery used its 2026 Global Innovation Conference to present the CheryMind vertical AI model, Ziwei chips and applications spanning ADAS, Robotaxi, robots and agriculture. The company says the same model, chip and engineering base can be reused across vehicle and non-vehicle scenarios, with Robotaxi deployment targeted for 2030 and agricultural service targeted for 2035.