BMW pulled back the curtain on a near-production version of its Neue Klasse technology at CES 2026, showcasing what will become one of the most significant updates to its in-car experience when it reaches dealerships later this year. The star of the show? A deep integration with Amazon’s Alexa+ smart assistant that transforms the BMW Intelligent Personal Assistant into a conversational AI capable of far more than basic voice commands.
I spent time with the system in a new iX3 at CES, experiencing demos from both BMW product manager Sarah Rau and Amazon engineering manager Sophie Zhou. The technology shows promise, though some implementation choices—particularly around the wake word—may frustrate owners.
A Conversational Leap Forward
Current BMW models already feature the BMW Intelligent Personal Assistant for handling calls, messages, and basic vehicle functions. This Alexa+ integration (even though it’s still an early version) represents a fundamental shift in capability. Built on large language models and generative AI, the system processes natural language requests in the cloud, allowing drivers to interact conversationally rather than memorizing specific commands.
The assistant activates by saying “BMW”—just the brand name alone, without requiring “Hey BMW” as a prefix. This design choice proves controversial in practice, as I’ll explain shortly.
During Rau’s demonstration, she put the system through a realistic road trip scenario. She asked it to navigate to Mount Charleston, add a CVS stop for sunscreen along the route, switch to sport mode, lower the cabin temperature, recommend Italian restaurants in Las Vegas, check table availability at Brezza restaurant, and play Las Vegas-themed music. The assistant handled each request successfully, though with noticeable processing delays between responses, mostly related to the spotty internet connectivity around CES and also the need to ping external API services, like Open Table.
Technical Architecture and Performance
The system operates on a hybrid model that relies primarily on cloud processing but includes local voice components for basic functions. When cellular connectivity drops, drivers can still control windows, HVAC systems, and navigate to previously saved locations using onboard speech recognition. Everything else—complex navigation queries, restaurant searches, music requests—requires an internet connection.
Amazon’s engineer explained that requests process sequentially rather than in parallel. When you ask the system to change driving modes and adjust temperature simultaneously, it confirms the first action before moving to the next. This approach ensures accuracy but introduces latency, particularly with multi-step commands. The CES demo environment’s fluctuating connectivity exacerbated these delays, though Zhou insisted normal operation should be noticeably faster.
The assistant maintains conversational context for approximately 16-17 turns or roughly one hour before clearing its memory. However, users can instruct it to save specific information—favorite destinations, family preferences, frequently visited locations—to long-term memory for true personalization across sessions.
Understanding Context and Limitations
One impressive aspect is how the system handles vehicle-specific constraints. During the demo, I asked the European-spec iX3 to close the windows, and it complied immediately. Rau noted that US-market vehicles, where regulations prohibit remote window closing, will have the assistant explain why it cannot perform that action—a significant improvement over current iDrive 8 and 8.5 systems that simply doesn’t provide context.
The assistant also demonstrates genuine understanding of natural language variations. When I asked about charging stations, saying “can you find me a charging station on the way,” it returned 20 options along the route. Requesting windows be opened “just a little bit” worked as expected. The system asked clarifying questions when needed rather than guessing at unclear requests.
It can even field knowledge queries about BMW products. When I jokingly asked whether the iX3 would get ventilated seats, it correctly responded that they’re coming later in the model’s lifecycle, citing product manager Mark Berger’s confirmation—information from a conversation I’d actually had with Berger two months prior, demonstrating the system’s ability to search current information.
Smart Home Integration Extends Functionality
The Amazon booth demonstration revealed capabilities beyond the vehicle itself. By linking an Amazon account, the system provides smart home control from your car. Zhou showed how a conversation started with Alexa in your kitchen—discussing a ski resort destination, for example—can continue immediately when you get in the vehicle. Simply saying “BMW, take me to the place we just talked about” starts navigation without needing to repeat the destination.
Along the route, the assistant can control home lighting, check Ring doorbell cameras for package deliveries or visitors, and manage any Alexa-connected smart home devices. Zhou noted they’re working on push notifications for Ring cameras so alerts arrive automatically rather than requiring drivers to ask.
The Wake Word Needs Work
Here’s where my enthusiasm hits a significant obstacle: the wake word implementation. The system activates on “BMW” alone, without requiring the “Hey” prefix. This means any casual mention of the brand name during normal conversation triggers the assistant. I raised this directly with both BMW and Amazon representatives, and they were aware of this feedback from media as well.
They also shared usage data showing customers naturally drop the “Hey” prefix anyway, preferring the shorter trigger. BMW initially tested “Hey BMW” but found users gravitated toward just saying the brand name.
I understand the reasoning, but disagree with the execution. Either provide full customization of the wake word—allowing owners to name their personal assistant—or mandate the full “Hey BMW” phrase to prevent false activations. The current implementation will frustrate anyone who discusses BMWs regularly during drives, and there’s currently no way to change it.
Real-World Use Cases and Learning Curve
The most frequently used functions, according to BMW, are navigation, phone calls, climate control, and driving mode changes. Customers also commonly ask about speed warning beeps—”what was that sound I just heard?”—which the assistant can explain and disable if requested. BMW will continue testing and refining based on real-world usage patterns.
When commands aren’t supported, the assistant now explains why rather than simply failing. Requesting an interior lighting change to a “Las Vegas” theme prompted a clear “that function is not supported” response.
Rollout and Availability
BMW will begin rolling out the Intelligent Personal Assistant with Alexa+ integration to vehicles running BMW OS 9 and BMW OS X in the second half of 2026. Initial availability targets the US and Germany, with additional markets following over time.
The integration comes standard with no additional subscription cost—you just need a basic Amazon account to access extended features like music streaming, restaurant reservations, and smart home control. Core vehicle functions (navigation, climate control, driving modes) work without any Amazon account, though you miss out on the conversational AI capabilities and third-party integrations.
Amazon’s representatives emphasized that because most processing happens in the cloud, BMW and Amazon can continuously add features and improvements after launch without requiring vehicle software updates. This cloud-based architecture should allow the system to evolve rapidly based on customer feedback and usage patterns.
