Source: Autonomous Driving Frontier, Public Online Materials
According to reports from Jiemian News Express and other media outlets, on October 13, Hello Robotaxi officially released its generative driving world model, HelloWorld. According to official introductions, the model supports describing requirements through natural language or directly editing vehicles, pedestrians, and motion trajectories to generate corresponding multi-view videos and LiDAR observations. From inserting crossing pedestrians to switching rain, snow, and day-night conditions, R&D personnel can construct scenarios around specific needs, expanding long-tail situations that are difficult to fully cover in real-world data collection.
Why is Hello building a world model? The answer lies in the timeline.
Zaofu Intelligence was established in June 2025, released its first vehicle model in September, achieved mass production and rollout in July 2026, and completed a USD 100 million financing round in September. In 15 months, it went from zero to mass-produced vehicle rollout. However, road test mileage, complex scenarios, and safety redundancy cannot be built solely with capital and speed. The world model is essentially Hello's way of leveraging generative AI capabilities to shorten this accumulation process.
HelloWorld: Where Does the Data Come From?
HelloWorld is part of Hello's "Daoyu" ten-billion-parameter large model system. The latter was jointly developed by Hello Robotaxi and Alibaba Cloud's Qwen team, covering cloud capabilities for data mining, world model simulation, and the VLA cloud driver foundation. HelloWorld takes on the role of data generation and simulation verification, providing scenario data for end-to-end model training.
According to public information, Hello Robotaxi's core technology team has a scale of over 500 people, with an average age of 27. It is characterized by an "AI-native" and legacy-free architecture, adopting a data-driven end-to-end large model approach. According to Hello's plan, by 2026, it will establish a high-quality dataset containing no less than 10 million clips. Given the limited real-world road test mileage, the generative world model has become a tool to fill the data gap.
This direction is not unique to Hello. According to public information, XPeng released X-World based on video diffusion generation technology, Xiaomi launched the Xiaomi Auto World Model, and Waymo built virtual environments leveraging Google DeepMind's Genie 3. Waymo stated that its world model can simulate extreme weather such as tornadoes, as well as scenarios where vehicles encounter wild animals.
Hello's characteristic lies in the tighter integration between its world model and its industrial shareholder structure. Alibaba Cloud provides the computing power foundation, CATL provides the electrified chassis and energy supply, and Ant Group contributes mobility operation scenarios and payment ecosystems. This four-party collaborative pattern is still rare among autonomous driving startups.
However, the uniqueness of the technical architecture cannot automatically translate into advantages in engineering implementation.
Mass Production Advancement and Safety Accidents
One month before the release of HelloWorld, Hello Robotaxi completed a new round of financing of approximately USD 100 million, with a post-investment valuation approaching USD 3 billion. The funds from this round will be used for a 10,000-GPU cluster computing power platform, end-to-end technology iteration for AI large models, and the construction of a high-quality data supply chain system.
From the establishment of Zaofu Intelligence in June 2025 to the mass production and rollout of the first OEM mass-produced vehicle model HR1 at the Dongfeng Nissan Huadu Plant in Guangzhou in July 2026, Hello Robotaxi took 15 months to complete the mass production and deployment path that took other players years. The HR1 is equipped with 8 LiDARs and 13 cameras, with a computing power exceeding 2,200 TOPS, and the current rollout scale has exceeded 100 units.
This speed is related to the business situation Hello faces. The growth rate of the shared two-wheeler business has slowed down. From 2018 to 2020, Hello's cumulative net loss approached CNY 5 billion, and the gross profit margin of the shared two-wheeler business in 2020 was only 6.7%.
In 2025, the Beijing Municipal Transportation Law Enforcement Corps filed an investigation into Hello for the illegal over-deployment of shared bicycles, and the CCTV "3·15" Gala exposed its electric bike rental business stores for illegally removing speed limits. Against this backdrop, Robotaxi carries not only the function of business expansion but also relates to Hello's new narrative in the capital market. In March 2025, its founder acquired the listed company Youon, which was seen as a key step in a "backdoor listing," and entering the Robotaxi sector is an important part of this plan.
Rapid advancement comes with safety challenges.
On December 6, 2025, a Hello L4 Robotaxi in Zhuzhou, Hunan Province, hit a pedestrian at a crosswalk, with one person being dragged under the vehicle. The accident occurred shortly after Hello Robotaxi launched commercial operations in Zhuzhou, directly leading to a complete suspension of its related businesses in Zhuzhou, Liyang, and other places. As of September 2026, the business in Zhuzhou has not yet resumed.
Chen Liteng, an analyst at the E-Commerce Research Center of NetSun, commented: "The business has quickly completed the setup from 0 to 1, but safety verification and scenario generalization capabilities have not yet matched the speed of commercialization. The internet-style rapid expansion mindset conflicts with the long-cycle and safety-heavy industrial attributes of the autonomous driving industry."
Scale Gap with Leading Players
Hello is not the first company to encounter the above problems. The challenges faced by latecomers will only be greater.
According to corporate disclosures, Baidu Apollo Go's fleet size reached 3,500 units by the first quarter of 2026, with 3.2 million fully driverless orders, and the total order volume increased by over 120% year-on-year. Pony.ai's global Robotaxi fleet size has exceeded 1,700 vehicles, and it has achieved positive profitability per vehicle in Guangzhou and Shenzhen. WeRide's global L4 fleet is about 3,400 vehicles, of which over 1,800 are Robotaxis, and the average daily orders per vehicle in China exceeded 21 in the second quarter.
Hello's HR1 currently has a rollout scale exceeding 100 units and plans to take the lead in carrying out normalized operations in Guangzhou and Shenzhen. Compared with leading players, there is still an order-of-magnitude gap.
At the same time, the Robotaxi economic model is improving. The material cost of Pony.ai's seventh-generation autonomous driving kit has dropped by 70% compared to the previous generation, and the target total vehicle cost for the 2027 version is set to drop to within CNY 230,000; the per-vehicle cost of Baidu Apollo Go's sixth-generation driverless car has been reduced to just over CNY 200,000. UBS predicts that in 2026, the per-kilometer operating cost of Robotaxis will drop to CNY 2.1, on par with traditional ride-hailing services.
Leading players have begun to verify the sustainability of the economic model, while latecomers still need to answer the question of cost structure. Yu Qiankun, co-founder and CTO of Hello Robotaxi, once stated that Hello has built a six-fold safety redundancy architecture and a dual-line data closed-loop system. Whether these investments can be absorbed in the per-vehicle economic model remains to be seen.
Guo Tao, a senior AI expert, commented that this is a typical latecomer catch-up strategy of trading capital for time. The business is still in the stage of technical verification and commercial trial operation. Safety accidents remind the industry that it cannot simply pursue the speed of expansion; the construction of technical stability and safety redundancy systems needs to be advanced synchronously with the deployment scale.
Safe Operation is the Key to Deployment
After the Zhuzhou accident, Hello Robotaxi's deployment preparations in multiple cities are still advancing. Cities such as Shenzhen, Changsha, Wuhan, and Hong Kong are promoting testing and deployment preparations, while Guangzhou and Changzhou have obtained road test permits. However, the approval for city access is becoming more prudent.
Chen Liteng pointed out that safety accidents have already affected regulatory trust, and city pilot approvals will be more prudent. Constraints such as less accumulated road test mileage compared to leading players, and insufficient complete vehicle supply chain and talent reserves, still exist.
The competition in autonomous driving ultimately depends not on who has the more dazzling technology releases, but on who can maintain a sufficiently low accident rate over a sufficiently long period. World models can generate a large amount of simulation data, but simulation data cannot replace complex interactions on real roads; the parameters of large models can be stacked, but the polishing of the safety redundancy system requires the precipitation of engineering experience.
Hello's industrial shareholder structure—Ant's scenarios and payment ecosystem, Alibaba's computing power and cloud services, and CATL's energy and chassis—provides it with unique resource synergy advantages. However, resource synergy solves the problem of "whether it can be done," but cannot solve the problem of "whether it can be done well."
The release of HelloWorld demonstrates Hello's technical execution capability in the field of generative AI. However, technical execution capability does not equal safety execution capability; the latter requires the accumulation of road test data, repeated verification of safety boundaries, and the ability to systematically reflect and improve in the face of accidents.
For Hello Robotaxi, the priority deployment plan in Guangzhou and Shenzhen in the second half of 2026 will be a critical test. Whether the HR1 fleet of 100 units can accumulate sufficient safe operation data in real urban environments, and whether the gap with leading players in the per-vehicle economic model can be gradually narrowed—the answers to these questions speak more to Hello's true position in this race than the technical parameters of HelloWorld.
The world model is a tool; safe operation is the true test.
- End -