Research Report on AI Applications in Cockpits, 2026

Description

AI Application in Cockpits: AI Services Become More Comprehensive, Convenient, and Refined.

In the first half of 2026, cockpit AI functions underwent initial upgrades across multiple dimensions, including from passive response to timely proactive action, from single-point functions to service loops, and from cloud-centric to edge-centric approaches. As agent functions are further enhanced, users' evaluation criteria for cockpit AI capabilities are also changing: the focus is no longer on which model achieves more advanced metrics, but rather on whose cockpit AI system can truly perform its tasks, protect privacy, anticipate user needs, and understand user demand.

Users' emphasis on actual experience and effectiveness is forcing cockpit AI to upgrade along three major lines simultaneously: more comprehensive, more convenient, and more refined.

More Comprehensive: From "Function" to "System"

In 2025, cockpit AI was positioned as a "function," focusing on different vertical scenarios and capable of completing single-shot services based on user instructions.

In 2026, the cockpit AI services of mainstream flagship vehicle models become more "systematic" and support the coordinated use of originally scattered non-safety AI application functions through central foundation models/agents to achieve a closed-loop service in vertical scenarios such as audio-visual entertainment, itinerary assistance, and local life services. Some cockpit products can even complete multiple tasks step by step based on independent planning, greatly broadening the scope of AI service scenarios, improving user experience, and laying the foundation for iterative upgrades of cockpit AI.

Especially, cross-domain integration at the AI level marks a more "comprehensive" technology foundation for AI services, and the layout of some OEMs has already been implemented:

For example, ZEEKR 8X's Super Eva fitted with cockpit-driving integration connects to the vehicle AI architecture centered on a WAM, supporting functions such as full-time, all-modal perception, deep thinking and decision-making, and full-domain scheduling (coordinating cockpit, driving, chassis, powertrain, etc.). It can also self-reflect and evolve, becoming more user-friendly with use.

In scenarios, Super Eva not only connects the in-vehicle and external ecosystems, enabling "speak-and-handle" daily tasks (e.g., ordering food, booking hotels, and processing work information directly via voice in the car), but also collaborates with G-ASD 4.0 to achieve autonomous driving and navigation, transforming from a "smart front passenger" into a "reliable driver," further expanding the scope of AI service scenarios.

Take IM's IM ULTRA AGENT 1.0 as an example. Through the IM FUSION NOVA cockpit-driving integration full-domain fusion intelligent architecture, the IM cockpit AI system allows for cross-domain linkage with IM AD ZETA and the fully wire-controlled Lizard Digital Chassis to realize functions such as changing destinations at will on the way. It can also be implemented in non-safety scenarios such as audio-visual entertainment, ecosystem interconnection, and personalized interaction to complete user command analysis and service closed loop.

More Convenient: Less Talk, More Action

Convenience hinges on how much effort users take for their desired outcome.

In 2026, users can feel the effects of AI applications more directly: with one less word, one less click, and one less second of waiting, the experience can reach a higher level. Therefore, cockpit AI in 2026 should further reduce the friction users encounter when accessing services: more direct interaction modes, fewer interaction steps and faster response speeds. Leveraging high-precision speech ASR technology, smarter AI algorithm scheduling and more human-centric workflow design, it minimizes redundant operations and page jumps, enabling multiple commands to be fulfilled with a single sentence.

Take the "picking up kid and navigating the way home" scenario for example:

Past: The user first clarifies the location A for picking up kid, then says "Start navigation"; the system asks "Where would you like to go?", the user replies "Location A, XX Road", and the system responds "Route planned for you", which means three dialogue turns are needed to complete one single task.

Now: The user simply says "Pick up kid", and the AI automatically fills in the destination and generates a route based on data stored in its memory, allowing three tasks to be completed via one vague voice command.

For example, Neusoft OneCoreGo 7.0 provides more comprehensive and convenient AI services through a "multi-in-one" sub-solution design. Multi-step operations of different application scenario functions can all be implemented by a single command through cross-agent collaboration technology.

One of the keys to realizing the convenience of cockpit AI is to implement multi-agent collaboration standard protocols and a unified scheduling framework; paired with an edge-cloud collaborative deployment environment, standardized agent communication, orchestration and execution protocols address interoperability challenges across cross-domain agents.

Extour Technology's MCP-Agent framework splits range detection, merchant screening, route planning, payment, etc. into separate agents. Different agents collaborate with each other through the MCP standardized protocol - for example, if a user says "order a low-fat coffee", the system can run through the entire link from product selection to ordering to navigation in a few minutes.

Leveraging context window optimization technology and memory modules, the MCP-Agent can continuously track successive changes to user requirements, such as adjusting coffee selections, cup sizes and pickup addresses during a coffee order, without requiring users to restate background information. Supported by standardized protocols for cross-service collaboration, it can process complex requests like "I will arrive at the office in half an hour, please recommend several low-calorie coffees" by automatically linking battery range detection, low-calorie merchant filtering, route planning and other services. All tasks can be completed with a single voice command, eliminating the cumbersome operation of switching between multiple independent applications in traditional solutions and significantly boosting the convenience of AI service interaction for users.

In contrast, Neusoft's NAGIC.AI solution also includes sub-agents for different scenarios. However, the complete multi-agent collaboration mechanism is achieved through the collaboration of modules such as Router, HCP, Memory, and Function Call (toolchain). The Router parses users' ambiguous intentions and dispatches corresponding scenario-specific agents. The Memory shares a unified memory pool to realize intention completion across different agents. Afterwards, the Function Call works with each agent to invoke underlying vehicle hardware, including navigation, ADAS, cockpit IVI, multimedia and other functions.

Furthermore, NAGIC.AI adopts a "distributed + centralized" solution. Based on standardized interfaces and a unified inference framework, it achieves layered adaptation to different computing power platforms (high-performance chips/mid-range platforms) and different systems (Linux/QNX/AutoSAR). It also includes built-in HCP (Heterogeneous Computing Platform) and AI Plugin Service Layer, providing standardized access and expansion capabilities for functional modules.

More Refined: Insight into "Implicit Demand as a Service"

The competition for the "refinement" of cockpit AI is unfolding from three levels - sharper perception, better understanding, and more measured actions. Wherein, sensing users' "implicit needs" is one of breakthroughs.

Users have diverse demand inside vehicle cockpits, ranging from "efficient commuting" and "relaxing" to "social interaction". Implicit needs in various scenarios need to be identified and fulfilled. In 2026, cockpit AI products typically process these implicit needs through a workflow consisting of perception, memory, comprehension, judgment, execution and verification. Vertical scenarios are pre-configured, and domain-specific agents are adopted to complete corresponding operations:

Taking perception as an example, cockpit AI is beginning to integrate vision, audio and vehicle signals. In limited scenarios such as "mobility services" and "child care", it can predict user needs in advance by sensing the occupant's expressions, body movements, blink frequency, steering wheel posture, etc., before the occupant issues voice commands, and provide end-to-end proactive services within a preset logical framework.

There are three types of scenario functions that OEMs may pay extra attention to, namely safety scenario functions, comfort scenario functions, and habit scenario functions:

Take Modelbest Technology's "SuperMate" as an example:

Modelbest Technology's design concept for cockpit AI is to replace "superposition of functions" with "extreme tacit understanding", and achieve "more restrained and restrained senselessness" through a closed loop of deep memory, real-time perception, situational understanding and proactive action. Typical functions include senseless car control, intervention of children's dangerous behaviors, accident status recognition and emotional comfort, etc.

Wherein, the most distinctive feature is the "active + senseless service" in the accident handling scenario of "SuperMate":

In addition, compared to other common in-cabin scenario functions, both SenseAuto and Neusoft Group have launched distinctive door open warning (DOW) functions. Such capabilities extend users' implicit safety needs beyond the cockpit to external road conditions.

For example, the "Safety Guardian" agent of SenseAuto, based on understanding capabilities of foundation models, achieves multi-dimensional risk identification, classifies and describes events such as dooring and car scratches, and through a safety closed loop and OpenClaw-based proactive warnings and real-time reminders, allows users to monitor the safety status of their vehicles anytime, anywhere, protecting their all-scenario driving safety.

Product Code: GX023

Definition

1 AI Application Scenarios in Automotive Cockpits

1.1 Status Quo of AI Applications in Cockpits
Characteristics of AI Cockpits
AI Application Scenarios in Cockpits: Status Quo
Evolution of Cockpit AI
1.2 Scenario 1: Voice Recognition
Development Roadmap of AI Foundation Models Integrated with Voice Recognition
Sub-scenario 1: Voiceprint Recognition
Sub-scenario 2: External Voice Recognition
Voice Interaction Suppliers Integrate AI Foundation Models
1.3 Scenario 2: Multimodal Interaction
Development Roadmap of AI Foundation Models Integrated with Face Recognition
Integration of Small Models in Lip Movement Recognition Scenarios
Integration of Small Models in Iris Recognition Scenarios
1.4 Scenario 3: IMS
Functional Implementation of In-Cabin Monitoring System
Development of AI in In-Cabin Monitoring Scenarios
Examples of AI Algorithms for In-Cabin Monitoring
In-Cabin Monitoring: AI Technology Applications by Chip Suppliers (1)-(4)
1.5 Scenario 4: HUD
Applications of AI Algorithms in HUD
1.6 Scenario 5: Radar Detection
AI Algorithms in Radar (1)-(2)

2 Status Quo and Trends of Cockpit AI Applications

2.1 Cockpit AI Market Data
Installations and Penetration Rate of AI-powered Voice Assistants, 2025
Penetration Rate of AI-powered Voice Assistants by Price Range, 2025
Penetration Rate of AI-powered Voice Assistants by Level, 2025
Penetration Rate of AI-powered Voice Assistants by New Energy Vehicle Type, 2025
Installations and Penetration Rate of Avatars, 2025
Penetration Rate of Avatars by Price Range, 2025
Penetration Rate of Avatars by Level, 2025
Penetration Rate of Avatars by New Energy Vehicle Type, 2025
Installations and Penetration Rate of AI Foundation Models, 2025
Penetration Rate of AI Foundation Models by Price Range, 2025
Penetration Rate of AI Foundation Models by Level, 2025
Penetration Rate of AI Foundation Models by New Energy Vehicle Type, 2025
2.2 Development Trends of Cockpit AI
Trend 1: Cockpit Assistants Spread from Chatbots to Agents
Trend 1: Key to Cockpit Agent Applications (1) - Multimodal Technology
Trend 1: Key to Cockpit Agent Applications (2)
Trend 1: Key to Cockpit Agent Applications (3)
Trend 2: Cockpit-Driving Integration Based on Unified Agent Foundation
Cross-Domain Integration AI Super Agent Layout of OEMs in 2026 (1)-(2)
Trend 2: Challenges for Cockpit-Driving Integration AI
Trend 3:
Trend 4:
Trend 5:
Trend 6:
Trend 7:
2.3 Resource Calculation for AI Technology Implementation in Cockpits
Resource Calculation (for Reference Only)
Comparison between Cockpit AI Assistant Applications of Major Foreign Brands
Advantages and Disadvantages of Different Cockpit AI Algorithms

3 Cockpit AI Application Cases of Suppliers

Overview of Cockpit AI Foundation Models from Suppliers
3.1 Huawei
Planning for AI Applications in Cockpits
Functional Construction of HarmonySpace
Celia's Voice Capabilities Based on Foundation Models
Celia's Perception Capabilities Based on Foundation Models
AI Functions of Harmony OS
Dynamics of HarmonySpace: MoLA Upgrade
Dynamics of HarmonySpace: Special Functions (1)
Dynamics of HarmonySpace: Special Functions (2)
3.2 Tencent
Cockpit System Upgrade (1): Evolution
Cockpit System Upgrade (2): Functions and Architectures
Foundation Models-Based Open Agent Platform
Cooperative Ecosystem
Intelligent Cockpit Foundation Model Framework
Foundation Models Enhance Interactive Functions
3.3 Alibaba
AI-Based Voice Scenarios
NUI Edge-Cloud Integration Platform Architecture
Functional Applications of Qwen Edge Foundation Models on IVI
Qwen Edge Foundation Models: Edge-Cloud Collaboration
Qwen Edge Foundation Models: Terminal Foundation Models
Qwen Edge Foundation Models: Application Scenarios
Qwen Edge Foundation Models: Application Roadmap of "Human-like Partners"
Qwen Edge Foundation Models: Next Steps
3.4 Baidu
Intelligent Cockpit Based on Ernie Bot
Multi-agent Collaboration Mode
3.5 ByteDance (Volcano Engine)
Next-generation Automotive AI Solution
Cockpit AI Assistant Construction
Cockpit AI Assistant Construction: Four Levels of Edge AI
Cockpit AI Assistant Construction: Functional Applications of Edge AI (1)-(8)
Cockpit AI Assistant Construction: Hardware Solution
Cockpit AI Collaboration: Suppliers
Cockpit AI Ecosystem: OEMs
3.6 Zhipu AI
Cockpit Design Architecture Based on AI Foundation Models
Scenario Design of AI Foundation Models
Design of AI Foundation Models for Cockpit Interaction Problems
Collaboration in Cockpit AI
3.7 SenseTime
Typical Functions of Intelligent Cockpits
Cooperative Dynamics:
Typical Cockpit Products: Edge Models and AIOS
Typical Cockpit Products: Edge Agents
Multimodal Processing Capability Framework
Multimodal Interaction Application Cases
In-Cabin Monitoring Products
3.8 iFLYTEK
Functions of Spark Model
Core Capabilities of Spark Model
Deployment Solution of Spark Model
Automotive assistant based on Spark Model
Functions of Spark Model: (1)-(6)
How Spark Cockpit Integrates AI Services
Application Technology of Spark Model
Full-Stack Intelligent Interaction Technology
Autonomous Vehicle AI Algorithm Chip Compatibility
Features of Multimodal Perception System
Spark Intelligent Cockpit 2.0
3.9 AISpeech
Functions of Automotive Voice Assistant
"1+N" Layout of DFM
Integrated Foundation Model Solutions
Multimodal Interaction Solutions of AI Voice Technology
Features of AI Cockpits
Cooperation Cases: (1)-(2)
3.10 Unisound
Automotive Foundation Model Solutions
Details of Foundation Models (1)-(3)
Applications of Shanhai Large Model in Cockpits (1)-(3)
Business Models of Automotive Voice Solutions
Voice Basic Technology
Edge-Cloud Collaborative Voice Technology
Cooperative Ecosystem
3.11 Pachira
Voice Foundation Model Solutions
Intelligent Cockpit Foundation Models (Hybrid Architecture + Open Fusion)
Applications of DeepSeek
AI Voice Solutions (1): Capability Types
AI Voice Solutions (2): Functional Scenarios
AI Voice Solutions (3): Features
AI Voice Solutions (4): Design Concept
3.12 Thundersoft
Foundation Model Layout
AI Cockpit Series (1)-(3)
AI Cockpit Software Foundation - Aqua Drive OS (1)-(2)
AI Cockpit Software Foundation - AI Box
3.13 Neusoft
Cockpit AI Solutions (1)
Cockpit AI Solutions (2)
Cockpit AI Solutions (3)
Cockpit AI Solutions (4)
Cockpit AI Scenarios (1)-(3)
Empowering Cross-Domain Integration Platform Products with AI
Applications of DeepSeek
Underlying Foundation of Cockpit Agents: OS and SOA
Next-generation Cockpit Mobility Solutions (1)-(6)
3.14 Desay SV
Main Application Scenarios of Cockpit Foundation Models
Multimodal Interaction of Cockpit Foundation Models
History of Automotive Voice Research
Overview of Voice Foundation Model Solutions
Voice Industry Solutions (1)-(4)
Dynamics in Cockpit AI Cooperation
Cockpit Development Trends
Upgrade of Smart Solutions
3.15 TINNOVE
Four Stages of Intelligent Cockpit Planning
AI Cockpit Architecture Design
AI large model application scenarios
TTi AI Cockpit (1)-(9)
3.16 PATEO CONNECT+
Voice interaction technology
Capabilities of Qing AI Voice (1)
Capabilities of Qing AI Voice (2)
Dynamics in AI Cockpit Product Cooperation (1)-(3)
3.17 Extour Technology
"1+3" Architecture of Cockpit AI Solutions
Edge-Cloud Collaborative Architecture of Cockpit AI Solutions
Underlying Architecture of Xinjie AI System
Typical Functions of AI System (1)-(4)
MCP-Agent Framework Accelerates MAS Intelligent Collaboration
3.18 Cerence
Core Voice Technology (1)
Core Voice Technology (2)
Voice interaction outside the vehicle
3.19 Horizon Robotics
Vehicle Intelligent Agentic OS
Cockpit-Driving Integration Chip for Vehicle Intelligent Agentic OS
3.20 Others
MINIEYE
LG
MediaTek

4 Cockpit AI Application Cases of OEMs

Foundation Model Applications of OEMs
4.1 NIO
Multimodal Perception Foundation Models: NOMI GPT
Multimodal Cockpit Interaction Applications Based on NOMI GPT
Applications of NOMI GPT in Cockpits
Intelligent Cockpit Functions Based on NOMI GPT (1)-(5)
ONVO Intelligent Cockpit Interaction Cases Based on NOMI GPT (1)-(2)
OnVO and NIO Offer Food Ordering Functions
4.2 Li Auto
Lixiang Tongxue: Building Multiple Scenarios
Lixiang Tongxue: Agent Architecture - Two Paths
Lixiang Tongxue: Ordering Scenario Analysis
Lixiang Tongxue: Payment Scenario Analysis
Lixiang Tongxue: Key Points of AI Master R&D
Lixiang Tongxue: CoT interpretability
Mind GPT (1)-(2)
Foundation Model Training Platform
Lixiang Tongxue: Foundation Model Capability Upgrade (1)-(2)
Lixiang Tongxue: Multimodal Interaction Cases
4.3 Xpeng
History of AI Foundation Model Applications
Intelligent Cockpit Solutions: XOS
Application Cases: (1)-(4)
4.4 Xiaomi
Automotive Foundation Models: MiLM
Voice Capabilities of XiaoAi Tongxue
Voice Task Parsing and Execution Process
Vehicle Recognition Functions (1)-(3)
Scenario Construction of XiaoAi Tongxue
Cases: Cockpit of 2026 New SU7 (1)-(4)
4.5 Leapmotor
Foundation Model 1.0
Foundation Model 2.0
4.6 BYD
Cockpit AI: From Independent R&D to Collaboration
Dynamics in Agent Collaboration
Application Cases of Xuanji AI Foundation Models in Cockpits
4.7 Geely
Xingrui AI Foundation Model
Architecture of Xingrui AI Foundation Model
Full-Domain AI System 1.0
Full-Domain AI System 2.0: Release
Full-Domain AI System 2.0: Architecture
Full-Domain AI System 2.0: Cockpit-Driving Integration Agent
Application Forms of Foundation Models in Cockpits
Flyme Auto Voice Interaction Capability
ZEEKR: Application Cases of Super EVA
ZEEKR: Agent Scenarios (1)-(2)
4.8 Chery
LION AI Foundation + iFlytek Spark Model + DeepSeek
Super Agent: Xiaoqi
AI Applications in Audio System
AI Cockpit Configuration in Vehicle Models
4.9 Changan
AI Model Matrix
AI Agent Matrix
SDA Intelligence System
Applications of AI Technology: (1)-(2)
Automatic Switching of Cockpit Scenarios and Functions
4.10 Great Wall Motor
Coffee Agent System: Application Scenarios
AI System Foundation
Improving the "Q&A Mode"
Agent Service System Based on Foundation Models
Upgrading the Overall Agent with a New Platform
WEY Equipped with Cockpit Agent
4.11 SAIC
Applications of IM Foundation Models in Automotive Voice
IM Foundation Model Application Case: IM L6
IM AI Foundation Models Build Active Perception Scenarios
DeepSeek Applied by SAIC
SAIC Super Agent Adopts a Full-Domain Fusion Intelligent Architecture
Application Scenarios of SAIC Super Agent
IM Agent Application Case: LS9
4.12 GAC
Intelligent Cockpit Solution: ADiGO SPACE
Applications of AI Foundation Models in Cockpits
Applications of DeepSeek in Cockpits
Edge-Cloud Integration Architecture of ADiGO Intelligence
4.13 BAIC
Development History of Cockpit AI System
Cockpit AI 2.0: Product Architecture of Baimo Huichuang
Applications of DeepSeek: Access to Baimo Huichuang
Cockpit AI Scenarios (1)-(2)
Cockpit AI Cross-Domain Fusion Era: Yuanjing AI
Three Levels of Baimo Huichuang AI Promoting Full-Domain Integration
4.14 FAW
Applications of DeepSeek (1)-(2)
Lingxi Cockpit: Implementing AI Scenario Services Through Full-Stack Independent R&D
Lingxi Cockpit: Application Scenarios of AI Functions
Lingxi Cockpit: Applications of AI Functions
4.15 Dongfeng Motor
Development of Cockpit AI System
Applications of DeepSeek: (1)-(2)
Cockpit AI System Based on "Tianyuan Architecture": From OS to Models, and Then to Cockpit Functions
Cockpit AI System Based on "Tianyuan Architecture": Foundation Model Application Architecture
Cockpit AI System Based on "Tianyuan Architecture": Cockpit Scenario Highlights
Cockpit AI System Based on "HarmonySpace": "Xiaoyao Cockpit 2.0"
AI Application: AI Agent of Voyah Courage
AI Application: Cockpit AI Technology Research Direction
Collaboration with Yan AI
AI Application Cases of Nissan
4.16 JAC
AI Cockpits: (1) - (2)
4.17 Tesla
Grok Cockpit Model Got Upgraded
4.18 BMW
Overseas and Chinese Cockpit AI Solutions
BMW Intelligent Voice Assistant 2.0 Based on LLM
4.19 Mercedes-Benz
MB.OS Digital World Delivers Personalized Services with MBUX Virtual Assistant
4.20 Stellantis
Automotive AI Applications (1)
Automotive AI Applications (2)
4.21 Others
Volkswagen Voice Interaction System Equipped with GPT
Toyota Cockpit AI Foundation Models Deployed