WeRide and Geely Farizon Push 2,000 GXR Robotaxis Into 2026
Rows of upgraded GXR robotaxis move from a Farizon assembly bay toward a mapped urban route, emphasizing the gap between factory output and live service.📷 AI-generated / Tech&Space
- ★The 2,000-unit plan is a 2026 delivery target tied to Q3 production of upgraded GXR robotaxis, not a 2024 rollout.
- ★GEN8 with Sensor Suite 8.0 adds a 600-meter thousand-line lidar claim, 17x higher point-cloud resolution, and more reaction time in high-speed scenarios.
- ★The hard part is the operating layer: local permits, fleet maintenance, charging, remote supervision, and reliable paid service.
FROM ASSEMBLY LINE TO STREET
WeRide and Geely Farizon are no longer pitching only another sketch of a future robotaxi. According to the official WeRide and Geely Farizon announcement, the companies are targeting delivery of 2,000 upgraded, purpose-built Robotaxi GXR vehicles in 2026, with the new model scheduled to roll off the production line in the third quarter. That corrects the stale frame in the current article: this is not a fleet promised by the end of 2024, but a live 2026 production and delivery plan.
The number is large, but it is not magic. A robotaxi business becomes real only when vehicles can be built, certified, serviced, charged, monitored, and dispatched without turning every city into a one-off engineering fire drill. WeRide reported 1,023 robotaxis in its global fleet as of January 2026 and expects the additional GXR vehicles to push its operating fleet past 2,600 this year. That is a serious jump. It is also a serious spreadsheet.
GXR is not just a slideware van. WeRide says the first mass-produced version, developed on Farizon's SuperVAN platform, launched in October 2024 and received approval in February 2025 for fully unmanned paid ride-hailing in Beijing. The company cites fully driverless commercial operations in Beijing, Guangzhou, and Abu Dhabi, while 2026 has added new external proof points: WeRide and Uber launched fare-charging fully driverless rides in Dubai, and WeRide with Grab opened public autonomous service in Singapore's Punggol district, initially with Safety Operators onboard.
The most interesting part of the announcement is therefore less glamorous than the word autonomous. Farizon and WeRide are trying to turn GXR from a converted vehicle into a factory-integrated robotaxi. The upgraded model uses WeRide's GEN8 system and Sensor Suite 8.0, while Farizon points to an AI-enabled drive-by-wire chassis, a mature supply chain, and a production system that should cut vehicle assembly from one hour to under ten minutes. The companies also expect total vehicle cost to fall by another 15 percent.
Ten minutes sounds like a marketing number, but robotics has a useful rule: the demo is not over until the service manual stops reading like a horror novel. Faster assembly matters only if calibration, redundancy checks, software configuration, regulator documentation, and routine maintenance can also survive months of rain, dust, curb strikes, and passengers carrying luggage.
The plan is not just a bigger fleet: GEN8 sensing, ten-minute assembly, and local permits will decide whether GXR moves from demo logic to daily transport.
A GXR sensor pod scans a rain and fog test corridor while a remote operations station monitors route readiness.📷 AI-generated / Tech&Space
SENSORS ARE NOT A PERMIT TO RELAX
The technical jump in GEN8 is concrete, at least on paper. WeRide lists a thousand-line lidar, 17x higher point-cloud resolution, a detection range of up to 600 meters, and more than 70 percent additional reaction time in higher-speed scenarios. It also claims stable perception in heavy rain or dense fog. Those are useful claims, but they remain company claims. Long-range lidar does not by itself solve a truck blocking a lane, a pedestrian appearing from behind a delivery van, or a local driving convention that only exists on rainy Mondays. The real world enjoys those details.
This is where a fleet differs from a product. One GXR can be an impressive demonstrator. Two thousand GXRs require an operating system: charging sites, cleaning, shift supervision, remote assistance, post-bump calibration, insurance, local maps, failure procedures, and a clear answer for who takes over when the autonomous stack asks for help. If that layer is missing, the vehicles become expensive parking-lot sculpture with excellent lidar.
WeRide is building that market layer in parallel. With Uber, it announced at least 1,200 robotaxis in the Middle East, spanning Abu Dhabi, Dubai, and Riyadh, by 2027 if regulatory approvals and performance milestones are met. In Singapore, the Ai.R public service is a deliberately bounded model where routes, operating hours, and Safety Operators are part of the design rather than a footnote. That is not weakness. It is how robotics gets introduced when the first mistake should not become the entire story.
Extra technical context came from WeRide's GTC 2026 appearance, where the company tied GXR to the NVIDIA DRIVE Hyperion platform and its HPC 3.0 compute unit. WeRide said the setup can reduce autonomous-driving-suite cost by 50 percent, cut total cost of ownership, and put current vehicle hardware costs around US$40,000, with a possible further 15 percent reduction as the fleet scales. Those numbers are promising, but robotaxi economics are not just sensor prices. They include depreciation, batteries, cleaning, supervision, empty miles, insurance, and local rules that do not care about a tidy cost curve.
If Farizon and WeRide hit the third-quarter production target, 2026 can become a useful test of factory-integrated robotaxis: not because 2,000 vehicles automatically transform transportation, but because that many vehicles begin to reveal what the system really costs outside a controlled pilot. The benchmark will not be the best promotional ride. It will be paid kilometers, uptime, local permits, interventions, maintenance, and whether a passenger can order the car without feeling like part of the beta program.
