The yardstick for ADAS is shifting from "can it demo" to production scale, data closed-loop efficiency, automotive-grade safety redundancy and compliant global expansion. Three 2026 Gasgoo Awards winners point to three distinct China paths. As intelligent driving moves from rule stacking and modular assembly toward end-to-end architectures, world models and Physical AI, the industry's evaluation criteria are shifting from "can it be demonstrated" to "can it be mass-produced, and can it reach global markets safely and compliantly." The Evaluation Standard Is Being Rewritten Intelligent driving is going through a technology paradigm shift. Early systems were built on a modular pipeline of perception, prediction, planning and control, relying on a large volume of hand-written rules that struggled to cover long-tail scenarios on urban roads. Since 2024, two-stage end-to-end has entered volume production, compressing perception and planning into neural networks while retaining some rule-based post-processing. In 2026, one-stage end-to-end — in which the entire chain from perception to control is neural-network-based — and world models have become the focus of leading players. Vehicles can understand physical laws, roll out trajectories and handle unseen scenarios inside generative simulation environments, an upgrade the industry sums up as autonomous driving built for Physical AI. Demand-side fundamentals are maturing. According to industry statistics, penetration of L2 and above driver assistance in the China market is approaching 70%, and urban NOA is moving down from models priced above RMB 300,000 into the RMB 150,000–200,000 mainstream band. Intelligence has become a core purchase decision. The data flywheel expands alongside penetration: the larger the production fleet, the richer the real-world data returned, the faster the model iterates, which in turn supports more design-ins — a closed loop of "production, data, iteration, more production." Companies that put their algorithms into million-unit fleets first are more likely to secure the training resources that are scarce in the end-to-end era. Technology evolution runs along three main lines. First, model architecture: from two-stage to one-stage end-to-end, with world models and reinforcement-learning models added to improve generalization at complex intersections, unprotected left turns and cut-in traffic. Second, vehicle-side engineering: most high-level solutions today run on domain controllers based on NVIDIA Orin-X and Orin-Y, and are migrating toward domestic chips and cockpit-driving integrated platforms, balancing compute, power and cost. Third, safety and compliance: L3-oriented solutions must build vehicle-level redundancy across perception, computing, communication, power supply, braking and steering. Overseas markets add further requirements such as Euro NCAP and the EU General Safety Regulation, making safety redundancy and regulatory fit a higher bar than raw algorithm accuracy. The competitive landscape now has three types of players on the same stage, and concentration keeps rising. One group is independent algorithm suppliers represented by Momenta, which serves multiple OEMs from a third-party position. Another is global Tier-1s represented by Bosch Smart Driving, which takes end-to-end capability developed locally in China and feeds it back into global programs. The third is L4-native companies represented by WeRide, which bring unmanned technology accumulated in Robotaxi down to L2++/L3 production and accelerate overseas expansion. The dividing line has moved from disengagement mileage in demo videos to four capabilities: production deployment scale, data closed-loop efficiency, automotive-grade safety redundancy and cross-border compliance. These four are also the coordinates for the three winners examined below. Momenta: The Independent Algorithm Supplier with the Largest Production Base Momenta received the "Top 100 Players of China's New Automotive Supply Chain (ADAS/AD)" award this year, with the R7 world model as its flagship product — an end-to-end solution positioned to "solve multiple scenarios with one model." The company follows a data-driven "two legs" strategy: scaling and accumulating data through its mass-production vehicle solution (Mpilot), exploring L4 operations through its fully driverless solution (MSD/Robotaxi, Robovan), and extending L4-grade safety foundations and cross-scenario generalization into volume production through the world model. Production volume is its most persuasive support. According to the company's 2026 interim report, cumulative installations of its production solutions exceeded 1 million units as of June 30, 2026, with about 321,000 units added in the first half, up about 83.7% year on year; it served 26 OEM customers, with 105 models in production and 219 cumulative design-ins, covering nine of the world's top 10 OEMs. In the same report the company states that it ranks first among third-party urban NOA suppliers in China. Third-party research estimates put its share of the domestic third-party urban NOA market at about 65%. A September 2026 People's Daily Online report, citing company information, offered an updated set of figures: cumulative installations above 1.1 million units, more than 110 models delivered, over 230 cumulative design-ins, with the latest 100,000-unit increment taking 36 days. It also disclosed that the company has started L3 development and plans to reach volume production within 2027 (planned). Note that the two data sets belong to June 30 and September respectively and should not be mixed. On the commercial side, revenue in the first half of 2026 was about RMB 1.602 billion, up 75.9% year on year, while adjusted net loss narrowed by about 96.6%, approaching break-even. The significance of R7 lies in entering mainstream models of luxury brands. According to official information, the all-new Mercedes-Benz long-wheelbase GLE launched in September 2026 comes standard with an urban and highway navigation assist system jointly developed by Mercedes-Benz and Momenta; the company says the system enables "driving everywhere" from launch, with OTA upgrades to the R7 world model and "parking-space to parking-space" capability to follow, making it the first luxury ICE model to carry R7. The Audi E7X is scheduled to upgrade to R7 within 2026 (subject to official timing), and Mercedes-Benz said during Auto Guangzhou 2026 that it will extend R7-based "parking-space to parking-space" to more models. The company also plans to launch its first production Robotaxi model fully equipped with the R7 world model in the fourth quarter of 2026 (planned). Image Source: Momenta Bosch Smart Driving: A Global Tier-1 Feeding Local R&D Back to the World Bosch Smart Driving (Bosch Suzhou) received the "Top 100 Players of China's New Automotive Supply Chain (ADAS/AD)" award this year, with "Bosch one-stage end-to-end high-level driver assistance" as its flagship product. As one of the world's leading automotive component and ADAS Tier-1 suppliers, Bosch in China follows a path of "local team leading development, China market launching first, then exporting back to the world" — a contrast to purely domestic algorithm companies. Its production portfolio is carried by the "Bosch Longitudinal and Lateral Driver Assistance" brand launched in 2025. The brand consolidates previously productionized solutions: the Supreme Edition addresses the SAE L2 urban driver assistance market; the dual Orin-X solution has been in production since late 2023 (initially highway navigation), with a two-stage mapless end-to-end function pushed over the air in early 2025; the Orin-Y-based solution entered production in the third quarter of 2025 and is evolving toward the award-winning one-stage end-to-end high-level solution; at Auto China 2026 in Beijing, Bosch showcased an L3 automated driving solution for the China market with vehicle-level redundancy. Globalization and compliance are its core strengths. Bosch's full-stack high-level ADAS helped the Zeekr 7GT earn a five-star Euro NCAP rating in 2026, among the first models to achieve five stars under the new protocols. Its first overseas mid-level driver assistance program entered production in the first quarter of 2026. In October 2026, Bosch and Chery signed a program nomination in Paris for Chery's first localized ADAS program in Europe, covering the fourth-generation multi-purpose camera (MPC4) and an integrated driving-and-parking system built on it, to be fitted to six all-new Chery models, with production planned for 2028 and total volume expected to exceed 800,000 units over five years. At global headquarters, Bosch's autonomous driving alliance with Volkswagen Group's CARIAD has completed joint development, and its AI software stack has been available for production programs since mid-2026. Image Source: Bosch WeRide: An L4-to-Production, Physical AI Company Taking Robotaxi Global WeRide won the Best Technical Practice Application Award this year for its "WeRide WRD 3.0 one-stage end-to-end driver assistance solution," officially classified in the ADAS/AD track. The company started with L4 Robotaxi and has since formed two growth curves — L4 driverless and L2++/L3 production driver assistance — reusing the Physical AI computing platform and unmanned safety capabilities built in the Robotaxi phase for scalable front-end production programs. WRD 3.0 is in a rapid ramp-up. According to the company's second-quarter 2026 results, WRD 3.0 shipments reached about 30,000 sets in the quarter, cumulative design-ins exceeded 30 models, and L2++/L3 revenue grew 219% quarter on quarter. Models already carrying the solution include Chery Exeed Sterra, GAC Aion N60 and Aion i60, and GAC Trumpchi Y7. Quarterly revenue grew 82% year on year and doubled sequentially, overseas revenue rose 164%, and the company was named to Fortune's 2026 "Change the World" list as the only autonomous driving company included; WRD 3.0 was also selected among the "Top 10 Automotive AI Applications" at the 2026 World Intelligent Connected Vehicle Conference. L4 expansion overseas is its distinguishing tag. The flagship L4 production Robotaxi GXR now operates in more than 12 cities worldwide, with a Robotaxi fleet of over 1,800 vehicles as of July 31, 2026, and fully driverless operations in Guangzhou, Beijing, Abu Dhabi and Dubai. In Dubai, the company works with Uber and the local Roads and Transport Authority (RTA), launching public passenger service in December 2025 and fully driverless paid operations in March 2026, becoming one of the first commercialized L4 Robotaxi services there (per company SEC filings). The extreme-scenario handling and safety redundancy accumulated in L4 operations form the basis of trust when the company exports its L2++/L3 solutions to overseas OEMs. WeRide's Robotaxi GXR in Madrid, Spain; Image Source: WeRide official WeChat account The Global Value of China Solutions Across the three winners, three identifiable paths for ADAS algorithm exports emerge, matching three types of global customer needs: Momenta offers a verifiable algorithm solution built on world models and million-unit production validation, in a combination of "end-to-end plus co-development with luxury brands"; Bosch Smart Driving leverages a global Tier-1's local R&D and worldwide safety compliance system to help Chinese models meet regulations such as Euro NCAP while landing local capacity; WeRide leads with overseas driverless Robotaxi operations and pulls its L2++/L3 production solutions into the Middle East and Europe. More fundamentally, the competitive focus has shifted: as end-to-end demos become standard, differentiation no longer rests on no-disengagement videos from launch events. Drawing on faster iteration cycles, a complete intelligent connected vehicle supply chain and million-unit production data, China's supply chain is turning safety redundancy, data closed-loop and cross-border compliance into a "China solution" with global replicability — reflected in three changes. First, competition is moving from algorithm demos to production engineering and the data flywheel. Automotive-grade compute, power and cost constraints mean that one-stage end-to-end and world models must run stably on limited hardware, while the data volume returned by million-unit fleets is becoming one of the more decisive barriers than single-point model parameters. China's leading suppliers hold a relative advantage in deployment speed and iteration cost. Second, value is concentrating in solutions that generalize across scenarios and allow responsibility to be defined. World models and vehicle-level redundancy let systems handle long-tail and extreme scenarios and reserve redundancy for L3 responsibility requirements; production-ready, verifiable safety capability is becoming the source of premium, rather than disengagement mileage in a lab. Third, overseas expansion is moving from "exporting features" to a systemic delivery of "safety compliance and localized operations." Euro NCAP ratings, EU regulations, data sovereignty and local operating licenses are the new thresholds; the luxury-brand co-development, Tier-1 global compliance endorsement and local Robotaxi commercialization described above correspond to three export paths that global customers can reference.