On October 16, 2026, Mercedes-Benz and NVIDIA announced an expanded strategic partnership to jointly develop next-generation driver-assistance and automated driving systems. Mercedes' vehicle operating system, MB.OS, will work with Hyperion, NVIDIA's in-vehicle computing and sensor architecture. Driving decisions will use the Alpamayo Stack, which includes reasoning capabilities, with safety mechanisms provided by Halos layered on top.
The two companies have already co-developed urban driver assistance for the CLA and have also announced a future Level 4 plan using the S-Class. With the expanded partnership, they say they will carry those plans into the next generation by combining AI-based decision-making with vehicle-wide safety design. NVIDIA's announcement and Mercedes' announcement both cite uses that include a future self-driving limousine.
What Moves Forward Beyond Urban Level 2 and the S-Class Plan?
The CLA's "MB.DRIVE ASSIST PRO," which Mercedes-Benz introduced on January 7, is an SAE Level 2 system that assists with urban driving from a parking space to a destination. Even though it uses NVIDIA's AI and driving software, the driver must still supervise. Being able to drive through a city is a different condition from not needing a human to be involved in driving. In its January explanation, the company said the system had been available in China since the end of 2025 and was scheduled to launch in the United States in 2026.
In the U.S. National Highway Traffic Safety Administration's (NHTSA) explanation of automation levels, Level 2 means the driver must keep paying attention even when steering and acceleration/braking are assisted simultaneously. At Level 3, the system handles driving, but a person takes over when requested. At Level 4, within a limited operating area the system performs the entire driving task, and occupants are not required to take part in driving. The numbers describe how the driving task is divided.
Lining up the two companies' announcements by this distinction and by status of availability clarifies where the latest news fits.
| Date of official material | Subject / plan | Stage described in the material |
|---|---|---|
| January 7, 2026 | CLA's MB.DRIVE ASSIST PRO | Urban Level 2 driver assistance. As of January, described as already available in China |
| January 29, 2026 | New S-Class Level 4-ready architecture | Design intended for future robotaxi operation via Uber |
| October 16, 2026 | Expanded partnership for next-generation systems | Joint development supporting driver assistance and future Level 4 uses |
The table classifies the subject and the development or availability status based on the CLA's official explanation, the S-Class's official explanation, and the latest joint announcement. The S-Class explanation is dated January 29, but the current page carries a note reflecting the September renaming of Hyperion.
The latest announcement is an expansion of a partnership built on the already published Level 4-ready S-Class plan; it is not an announcement that a Level 4 service is launching.
Between a "Level 4-ready" design and a service that carries passengers without a driver, implementation in vehicles and the definition of operating conditions remain. The October announcement does not name the models to be sold, the cities of operation, or a service start date. Nor can it be read as a promise that existing CLA or S-Class vehicles will become Level 4 through over-the-air updates.
Parallel Safety Mechanisms Supporting Reasoning AI
The design NVIDIA describes for the S-Class lets AI make driving decisions while a separate, conventional software stack runs in parallel. It also makes the computers redundant and uses radar and LiDAR in addition to cameras. Different means are built in for software decisions, for the sensors that perceive the surroundings, and for the hardware that keeps computing.
For the CLA's urban driver assistance, Mercedes-Benz said it uses about 30 sensors, citing 10 cameras, 5 radars, and 12 ultrasonic sensors. The Level 4-ready S-Class plan, by contrast, emphasizes sensor diversity including LiDAR and redundancy so that operation continues even if part of the computing fails. The difference between the two cannot be explained by AI model performance alone. Assistance that a human supervises and operation that needs no human involvement in driving require different provisions on the vehicle side.
In this partnership as well, each technology plays a different role. MB.OS supports driving functions and over-the-air software updates, while Hyperion combines in-vehicle computers and sensors. The Alpamayo Stack serves as driving software that reasons about situations, supporting point-to-point urban driving assistance, automated parking, and preventive safety. Halos adds a mechanism, including safety software that runs in parallel, to Mercedes' own safety foundation.
NVIDIA's official explanation of Halos describes using an AI-based driving pipeline together with conventional software that separates processing by function. It covers everything from the design stage through validation, and also incorporates simulation and safety functions. The ability of AI to choose actions in complex situations and the mechanism for keeping and verifying that behavior within safe bounds are treated as separate responsibilities.
These are, however, designs as described by the companies. The joint announcement does not include any guarantee that the parallel software will always catch AI errors, nor results from independent testing showing how much accidents would be reduced in these vehicles.
Another distinction worth drawing is between publicly released AI models and the software that actually goes into a vehicle. On August 4, NVIDIA announced that its open model "Alpamayo 2 Super" is available for commercial use. In the development path the company describes, a large model runs in the cloud and produces data that records reasoning processes as well as synthetic training data. That serves as a teacher for "distillation," which transfers capability to a smaller model that is then optimized to run in real time in the vehicle.
With this path, the large model used during development and the model that runs efficiently in the car can be kept distinct. But the October announcement does not disclose the specific name or scale of the model to be installed. From "using the Alpamayo Stack," one cannot conclude that the public Alpamayo 2 Super will run in the car as is. Turning reasoning capability into a production-vehicle feature requires implementation and validation matched to each vehicle's sensors and computers.
A Shared Foundation Extends to Volvo; Next Is Defining Operating Conditions
The day before the Mercedes-Benz announcement, on October 15, Volvo Cars also announced an expanded collaboration with NVIDIA. What it said it will use in its next-generation vehicles is likewise Hyperion, the Alpamayo Stack, and Halos OS. Covering highways, urban streets, and parking, it outlined a plan to introduce, in stages, features that let the driver stop watching the road ahead continuously.
Taken together, the two announcements suggest that what NVIDIA offers multiple automakers has broadened from in-vehicle computing power to a foundation for driving software and for safe development and validation. Vehicle makers combine it with their own vehicle design and safety technology, while NVIDIA supplies technology that can be used in common. The fact that Mercedes' announcement explicitly names its own MB.OS also points to this division of roles.
Even so, adopting the same technology names does not mean cars from different makers will drive the same way in the same places. Beyond sensor configuration and implementation in the vehicle, the conditions under which a function can be used, including which roads and what weather, will determine actual services. Whether data will be shared between makers, or whether contracts are exclusive, cannot be judged from these announcements.
In the next Mercedes-NVIDIA announcements, how concrete the target vehicles and operating conditions become, along with validation results and operating arrangements, will be the basis for judgment. Once those conditions are shown, users will be able to compare urban driver assistance that requires continued supervision with automated driving services in which they ride as passengers, as functions they can actually use.