By 2026, one of the most significant shifts in the automotive market will be the widespread adoption of large AI models in nearly all mainstream new cars. According to current data, over 90% of new energy vehicles priced above 200,000 yuan in the domestic market now come equipped with in-car large models, while the penetration rate in the 100,000-200,000 yuan segment has also surpassed 65%.

Newer automakers like XPENG, Li Auto, and Xiaomi have all developed their own large models from scratch. Traditional automotive giants such as BYD, Chery, and Changan have also embraced collaboration with tech companies, striving to tailor general-purpose large models for automotive applications. It’s safe to say that the era of one large model per car has officially dawned.

However, the only remaining concern amid this surge of automotive AI is that consumers are largely unimpressed.

Survey data from JL&F Consulting reveals that smart cockpits rank only ninth among factors influencing users’ car purchase decisions. Most users are unwilling to pay extra for in-car large model functions. Simply asking around among friends and family reveals that few people have strong opinions about large models in cars, and many are unaware of the differences between these models and previous in-car voice assistants.

So, why are automakers so enthusiastic, only to face a lukewarm response from consumers?

Behind this indifference lies a profound mismatch between technological advancements and user needs.

Tesla’s Gale, Crayfish’s Wave

Let’s first explore why automakers are uniformly eager to promote large models in cars. Objectively, this is not a technological upgrade driven by consumer demand but rather a forced response to industrial and technological trends.

The so-called industrial trend was most directly triggered by Tesla’s introduction of Grok in its vehicles in the summer of 2025. Frankly, when Chinese automakers develop new energy vehicles, they often look to Tesla for guidance. The mantra is: if the competitor moves, we follow; if they don’t, we’re unsure what to do. Once Tesla validated the value of large models in cars, automakers naturally followed suit, launching their own “intelligent travel companions.”

The technological trend stems from the fact that AI tools are indeed on the path to improvement and maturity. The mobile phone, PC, and IT markets are all eager to implement AI, and the automotive industry is no exception. Especially this year, the wave stirred up by crayfish (a metaphor for the AI craze) has been enormous, causing anxiety among companies fearful of being left behind. Since cars are naturally multi-system integrated terminals suitable for intelligent interaction, accelerating the integration of large models in cars is essential to ensure they remain competitive in the race for intelligent entities.

As a result, by 2026, all new domestic cars priced above 200,000 yuan will be equipped with large models; those in the 100,000-200,000 yuan range will be promoted as offering affordable in-car large models that are absolutely worth the price.

The industry and the era are pushing automakers to accelerate the integration of large models in cars. However, amid this bustling scene, the voices of users have been overlooked.

Undisturbed Is the Greatest Kindness

Why are users’ voices not being heard? Because large models talk too much, leaving users unable to get a word in.

Especially when listening to a large model prattle on while driving, the experience can be overwhelming. Consider this: when we engage in voice interaction while driving, we obviously hope for precise and rapid completion of the interaction without disturbing our subsequent driving operations. However, large models are inherently flawed in this regard. Their style is natural dialogue, with content that is detailed and somewhat verbose.

In short, in-car large models are much more talkative than previous in-car voice assistants. When driving, relatives and friends try their best not to disturb the driver, but large models seem to enjoy doing the opposite by constantly outputting information. For example, when a car owner says to navigate to a certain location, an operation that should take 1 second, the large model is likely to ask several questions, such as whether to stop somewhere along the way or whether to choose a more scenic route, mainly adding unnecessary complexity and verbosity. It’s as if you’re in a scene from “A Chinese Odyssey” with the Tang Monk’s incessant chatter.

What’s worse is that large models not only respond verbosely but also have weak boundaries. They are often accidentally awakened or operated, inexplicably joining in-car conversations. Turning them off requires additional effort, and by then, you’ve forgotten what was being discussed in the car.

Some in-car large models with advanced scenario-based capabilities, which come with hefty subscription fees, take disturbance to the extreme. These models identify scenarios and take proactive actions, such as opening windows for ventilation and playing music when they think the driver is fatigued, regardless of whether the driver is genuinely tired or just has small eyes, whether it’s freezing outside, or whether the music might startle the driver and cause danger. They act on their own, thinking, “I’m AI, so listen to me.” When they detect traffic congestion ahead, they constantly recommend rest areas and parking lots, regardless of whether the driver is late for work, simply suggesting you take a break.

Surveys by ZaoZi Auto Research show that over 60% of car owners say that in-car large models can only execute simple commands with fixed phrasing and often fail to respond accurately or execute commands when faced with slightly more complex multi-intent or long sentences.

Perhaps these issues are temporary and can be overcome, but the costs should not be borne by consumers.

After all, what consumers need is simply to be left alone by AI and to drive in peace.

Too Many Jarvises, Not Enough Humans

Even if consumers are not frequently disturbed by large models, they face a new problem: What exactly is the point of having a large model?

As automakers continuously promote in-car large models as the most versatile personal assistants, the Jarvis we’ve been eagerly awaiting, consumers are simultaneously discovering that there are just too many Jarvises, and they seem a bit excessive.

In real-world usage scenarios, consumers often find themselves in a dilemma when faced with in-car large models:

While driving, attention must be focused on the road. The jokes, copywriting, and travel guides that automakers tout are absolutely useless in this context. Think about it: who, in their right mind, would already be driving and then ask the in-car large model to write a travel guide on the road? As for functions that might be frequently used while driving, such as navigation and answering calls, traditional AI voice assistants already cover these adequately, making the “Jarvis powers” of large models unnecessary.

More critically, these large models emphasize “natural language understanding and fuzzy intent recognition.” The term “fuzzy intent” is quite alarming in driving scenarios. For example, if a user says, “I’m hot,” intending to adjust the air conditioning, the large model might misinterpret this as the user wanting to open the windows. If this happens on the highway, it could pose a direct driving risk.

Then, when parked and free to use the in-car large model without restraint, users are about to issue a command when they find themselves instinctively reaching for their phones. Isn’t the large model that operates smoothly, has a complete ecosystem, and is familiar with my usage habits just lying quietly in my phone? Why bother using a crippled version on the car’s infotainment system?

You Fragile Little Thing

If in-car large models are useless, that’s one thing, but the key issue is that they are often not user-friendly.

As mentioned earlier, due to limited on-device computing power, many in-car large models are crippled versions that are not user-friendly. To provide users with a full-fledged large model experience, many automakers have also launched cloud-based versions. This is all well and good, as automakers admit their cars have limited computing power, but they forget that their network capabilities are also lacking.

When in-car large models require a stringent network environment, users will find that once they encounter scenarios like underground parking garages, remote roads, or highway tunnels, the large model loses signal and directly malfunctions. What’s even more frustrating is that before large models were introduced, at least the in-car voice assistant was still usable. Now, with a network outage, even basic voice commands cannot be executed. The good news is that large models have enhanced in-car AI capabilities; the bad news is that they are also more prone to failure. What kind of hellish joke is this?

Another fragile aspect of in-car large models is what's known as "OTA betrayal." Many users have found that the large models they use become worse with each upgrade. They started out quite smooth, but after a few OTA updates claiming to add new features, the originally useful commands can no longer be recognized, and frequent lagging and freezing occur.

Consumers say, “I’m buying a car, not chasing an idol in the Chinese entertainment industry. I don’t need this sense of fragility.”

Aren’t Large Models Supposed to Be Free?

After all is said and done, the biggest gripe about large models in cars is that their differentiation is too small.

There is little difference between large models from different automakers; there is no difference between in-car large models and those on phones, with the latter offering a better experience and being completely free; even more infuriating is that often, there is little difference between in-car large models and older in-car AI voice assistants. These “similarities” leave users with no reason to like them, let alone the motivation to pay for them.

More often than not, consumers use in-car large models for a couple of days after buying a car and then never again. This scene is reminiscent of those useless digital products we’ve bought and the AI software we’ve installed on a whim. But the problem is that cars are expensive enough to require thorough purchase decisions. No matter how technologically advanced or in line with the times a product is, if it’s useless to us, it only elicits a shrug.

Automakers say, “Large models are in cars now. Come buy a car!”

Users respond, “Your message has been received. The system is currently processing your request. The estimated wait time is approximately 3 to 5 years.”