Preface Competition in the automotive industry has entered its second half, and the development of vehicle intelligence is reaching a critical watershed. If the past few years were the "fever phase" of exploding intelligent technology and rapid feature proliferation, then 2025-2026 marks the point where intelligence in China's cars has officially moved into the deep waters of structural adjustment and value restructuring. Autohome has published its "China Intelligent Vehicle Development Trends" report for eight consecutive years, staying in step with the evolution of intelligence in China's passenger-car market. Drawing on Autohome's big data, its comprehensive model database and years of continuous user survey data, we have identified four core changes in the intelligent development of China's new-energy vehicles. We also find that so-called "equal intelligence for petrol and electric" has not yet arrived, and that robotaxis are going through visible growing pains. 01 New-Energy Intelligence: Deep Adjustment, Entering the Second Half Trend 1 - Breaking out of the red ocean: the survival battle for intelligent cars amid homogenisation Intelligence levels in the new-energy market have risen fast. Smart-cockpit penetration passed 90% in the first half of 2026, yet penetration at the individual feature level has polarised. On one hand, functions such as dual-zone and above voice wake-up, 12.3-inch central screens, app stores and Wi-Fi have all passed 80% penetration and are now fully standard. On the other, a group of features sits below 30% - newer items such as the sky screen and exterior voice interaction, but also features that have been around for a while without ever gaining real traction, such as facial recognition and AR-HUD. The result is that NEV smart-cockpit hardware splits cleanly "in two": very few features fall inside the 30%-80% band, leaving an obvious empty zone. Intelligent-driving penetration shows no such fault line. Instead, growth has diverged. Basic functions such as ACC adaptive cruise and autonomous emergency braking have reached roughly 70% penetration but have stopped growing, while parking features such as memory parking and traced-path reversing have become the new growth engine. Yet as intelligent features spread, heavy homogenisation has become a challenge every carmaker must face. Mainstream cockpit features are standard on more than 80% of cars above RMB 100,000, and intelligent driving on more than 80% of cars in the RMB 200,000-plus mid-to-high-end segment. That means the configuration arms race is loading cost pressure onto carmakers without delivering any differentiated product advantage. Intelligent development has entered its second half, and the battle to break out of the red ocean has begun. Trend 2 - Value restructuring: cockpits move from stacking features to elevating experience Cockpit intelligence is being elevated on three fronts: interaction, sight and sound, and comfort. In interaction, voice remains the core way to control the car apart from touch, and its capability keeps expanding. First, more functions can be triggered by voice - more than a third of models now support six or more voice-controlled functions. Second, zone wake-up reaches further back: penetration of second- and third-row zone wake-up is 41.4%, standardised first in the mid-to-high-end market. More importantly, with large AI models in the loop, user satisfaction with voice interaction has improved. Basic complaints such as "interaction feels unnatural" and "functions are limited" have eased, and "not comprehensive enough in handling requests" has jumped to become the number one source of dissatisfaction. User expectations have moved from "it works" to "it works well". A quarter of new-energy cars can already be operated in two or more ways beyond touch, mostly voice plus gestures or facial recognition. But because facial recognition still has functional gaps in multimodal interaction, almost no model can trigger actions cooperatively across modalities. So what passes for "multimodal interaction" today is essentially voice plus gesture control - and with gesture control itself still at low penetration, true multimodal interaction has only just begun: combined penetration of the two functions is just 6.3%. In sight and sound, screens getting "bigger" and "more numerous" remains the main theme. Central screens above 12.3 inches are already standard in the new-energy market and are now growing further, moving quickly past 15 inches from 2024; 15-inch-plus screens are standard in the RMB 200,000-500,000 market. Even larger screens above 20 inches penetrate more deeply in the RMB 200,000-400,000 band, overtaking the RMB 400,000-plus premium segment. The RMB 300,000-plus premium market is also the main battlefield for multi-screen entertainment, with passenger-side entertainment screens and rear multimedia screens going mainstream quickly. What users want from screens is shifting from "entertainment" towards "driving": AR-HUD and central sky screens that serve the driver's seat and display driving and vehicle information are in strong demand. Expectations for entertainment screens have risen too - demand for multi-screen linkage is strong, and plenty of users want high resolution and a strong audio-visual experience. After "more screens" comes "smooth to use, enjoyable to watch". In comfort, second-row zero-gravity seats have grown explosively in the RMB 300,000-plus premium market, reaching close to 50% penetration in the RMB 400,000-500,000 band. Needs differ hugely by seating position: front passenger and second-row occupants chase maximum comfort and want zero-gravity seats and seat massage. We also found that buyers with budgets below RMB 200,000 actually rate comfort higher than those spending more, yet penetration of both zero-gravity seats and massage is still very low at that price. "Comfort above its class" could become a differentiating selling point in the sub-RMB 200,000 red ocean. Trend 3 - Alternating heat and cold: advanced intelligent driving keeps spreading, but the experience cools The "intelligent driving for all" push launched in 2025 carried into 2026, yet in the sub-RMB 100,000 market penetration of L2 and above has barely moved. The low-price "equalisation" carmakers promised has not materialised, constrained by cost. Above RMB 100,000, however, L2 and above has clearly levelled up: advanced intelligent driving with highway NOA and urban NOA has replaced basic L2 as the absolute mainstream. The RMB 200,000-400,000 band is where advanced intelligent-driving penetration is most pronounced, even ahead of the RMB 400,000-plus bracket. Assisted parking and remote parking also jumped sharply above RMB 200,000, becoming essentially standard in 2025 and growing only slightly in 2026. Growth in parking functions in 2026 has concentrated on memory parking, which broke through 50% penetration in the brutally competitive RMB 300,000-400,000 band and went mainstream first. Even so, penetration still lags demand. Highway and urban NOA penetration remains below what users expect - and since these are the two most-wanted intelligent-driving functions, advanced intelligent driving across every price band should stay the direction carmakers push. By price band, assisted parking is the function worth improving first below RMB 200,000, while buyers above RMB 400,000 prefer the sense of occasion and convenience that remote summoning brings. Competition is no longer confined to L2. Since late 2025 a number of carmakers have obtained L3 access or road-test permits. Based on the hardware fitted to approved models, we set out a rough baseline hardware specification for entry-level L3. Measured against it, more than 40% of new models launched above RMB 300,000 in the first half of 2026 already meet L3 hardware requirements - "forward-looking redundancy" that carmakers are building in as a guarantee of future function upgrades. In contrast to this fierce hardware race, user experience is cooling. Satisfaction with intelligent assisted driving has not improved in two years, and there is no significant difference between price bands. By usage frequency, the share of heavy highway-NOA users has grown, but the share of heavy urban-NOA users has fallen sharply. The core reason users avoid these functions is technological maturity: 66% still have safety concerns about assisted driving in complex situations such as cut-ins, pedestrians or vehicles appearing suddenly from behind an obstruction, and abrupt braking. Precisely because the experience falls short, users are more cautious about L3 and L5. The share who believe fully autonomous L5 will arrive before 2035 has fallen compared with 2024. The wait-and-see attitude is hardening. Trend 4 - Hardcore technology plus visible premium defines the new premium Against a backdrop of intelligent homogenisation, emerging features with low penetration are the ones most likely to stand out and become levers for differentiation. These features play different roles in the purchase journey. Safety- and experience-boosting functions such as driver-incapacitation assist and ultra-fast charging actively encourage purchase and convert well at the point of sale. More interactive features such as facial and voiceprint recognition and the sky screen attract attention in communications and work better as customer-acquisition hooks. Rear dimming privacy glass and intelligent air suspension do both. Among these, rear dimming privacy glass and the sky screen currently sit at just 0.2% penetration, giving them the strongest differentiation potential. Beyond differentiation, intelligence is also redefining "premium": emerging features with visible premium appeal carry more premium character. Rear multimedia screens, zero-gravity seats and AR-HUD are noticed the moment a user steps into the cabin, while cutting-edge, hardcore chassis technology such as magic-carpet intelligent suspension more readily earns user recognition. These features therefore penetrate more deeply in the RMB 400,000-500,000 band, where premium positioning is pronounced. Emerging features do not always enter at the top, though. The RMB 300,000-400,000 market has a stronger scale advantage than pricier segments and has instead become the testing ground for new kit. Traced-path reversing and memory parking, for example, exceed 50% penetration in the RMB 300,000-400,000 band, more than 20 percentage points ahead of higher price brackets. Among interaction features, facial recognition and gesture control, along with cabin-atmosphere items such as active ambient lighting and active noise cancellation, likewise penetrate more deeply there than above RMB 400,000. Fast charging behaves differently: charging to full in under 20 minutes actually penetrates more deeply in the RMB 100,000-300,000 market, so "fast-charging equalisation" is arriving first there. Households dominate that band and feel refuelling anxiety most acutely, so carmakers build fast replenishment into their core pitch. Fast charging at these prices is also achieved mainly through fast-charge-capable cells - an 800V high-voltage platform is not a prerequisite - and falling battery costs make equalisation possible. In the premium market, charging speed is not a must-have selling point, but the more extreme standard of charging within 15 minutes has the highest penetration above RMB 500,000, retaining a premium character. 02 Petrol-Car Intelligence: A Generational Gap With New Energy Remains In the first half of 2026, smart-cockpit penetration in petrol cars was 63.3%, against more than 90% for new-energy vehicles - still a wide gap. In terms of cockpit fitment, though, petrol cars are actively embracing intelligence: the share of new models launching with a smart cockpit has climbed steadily from 50% in 2023 to 70% in the first half of 2026. By price band, NEV cockpits have achieved equalisation across the board with penetration above 80% at every level, while petrol cars remain an intelligence lowland below RMB 100,000 at just 34%. The RMB 100,000-200,000 and RMB 400,000-500,000 segments even declined in 2026, leaving overall petrol cockpit penetration growth short and stuck in the 60%-70% range. The RMB 200,000-300,000 band has instead become the most fiercely contested ground for petrol cockpit experience: whether voice zone control, large central screens or passenger screens, penetration there leads. Petrol-car intelligent driving is developing steadily, with both penetration and standard-fit rate breaking 55% in the first half of 2026. Overall it sits at the stage of "mainstream adoption largely achieved, full standard fitment still some way off". By price, petrol cars below RMB 100,000 almost never offer assisted driving; above RMB 100,000 only the most competitive RMB 200,000-300,000 band approaches standard fitment at 80% - a two to three year generational gap behind new-energy cars. That high penetration is driven by Volkswagen, Toyota and their peers, traditional makers working with suppliers to catch up on intelligent driving in this profit heartland as a defence against the technological onslaught of new-energy rivals. Splitting out L2, highway NOA and urban NOA makes the gap even clearer. Excluding the sub-RMB 100,000 market, new-energy cars have entered the popularisation phase of advanced intelligent driving, with highway/urban NOA penetration around the 50% mainstream mark at every price band. Petrol cars, by contrast, have not broken 80% penetration for L2 and above at any price band and are still in the phase of making basic intelligent driving standard. The gap in parking intelligence is far wider than in driving intelligence. Petrol cars concentrate on assisted parking, fitting it more quickly in the RMB 300,000-plus premium market, but remain at an early stage with low penetration for advanced functions such as memory parking and remote parking. Even with that gap, users acknowledge how far petrol cars have come. Nearly 90% perceive that petrol-car intelligence has improved over the past two years; pain points have eased and the perception that "petrol cars mean weak intelligence" has reached an inflection point. Petrol-car buyers also want intelligent driving badly. There is a gap versus new-energy buyers, but current fitment still sits well below expectations - and the gap between expectations for advanced intelligent driving and actual penetration is wider still, making it the key threshold for petrol cars to break through on intelligence. 03 Robotaxi: No Profit Inflection Point Yet, Experience Growing Pains Stand Out In 2026 the global robotaxi industry is scaling up and spreading across more cities, but the money still is not coming in - and the players keep raising their stakes. In March 2026 more than 100 Apollo Go vehicles in Wuhan were grounded at once, triggering a safety review across China's entire robotaxi industry and a suspension of new operating permits. The regulatory turbulence triggered by safety incidents is becoming a hidden cost that pushes the industry's profit window further out. On the user side, after two years of operation and promotion robotaxis have not broken into the mainstream. Only 11.7% of users have ever tried one, up just two percentage points on 2024 - still confined to a handful of enthusiasts and a few pilot districts. Worse, satisfaction among those who have tried one has fallen sharply: 76% of users who tried a robotaxi in 2024 rated the experience good, a figure that dropped to 66% in 2026. The industry urgently needs to shift from flexing technical muscle to doing the hard work on service experience. Conclusion Intelligent development has entered its second half, and competitive differentiation is now inevitable. Whether to keep pushing intelligent driving down the price scale and deepen the leap in experience, or to build differentiation and premium models around emerging features, is a question every carmaker will have to keep working on. The Autohome Research Institute will stay data-driven and user-centred, continuing to track the intelligent development of China's auto market. We hope to give manufacturers strategic reference and a decision-making basis for navigating the intelligent homogenisation cycle and finding new anchors of value. (By the Autohome Research Institute)