Shanghai, China – February 8, 2025 – Global mobility technology company ECARX (NASDAQ: ECX) has successfully completed the rapid deployment and deep integration of the DeepSeek-R1 model into its in-vehicle AI large model application, ECARX AutoGPT. This marks a significant upgrade following its debut at CES 2025 in Las Vegas. The integration not only expands AutoGPT's on-device processing capabilities but also enhances its inference, decision-making, and specialized domain expertise, once again pushing the boundaries of intelligent automotive interaction.
ECARX AutoGPT is a proprietary in-vehicle AI model built by ECARX based on general large language models, tailored for mobility scenarios. It integrates four core capabilities: AutoAgent AI Intelligent Agent, AutoFlow AI Task Execution, AutoScene AI Scenario Engine, and AutoEco AI Service Ecosystem. With natural language interaction, autonomous understanding, multimodal perception, task planning, memory, and tool utilization, AutoGPT can automatically execute complex tasks. Additionally, it features a comprehensive global AI service ecosystem centered around AI agents, enabling partners to deliver rich and practical AI-native applications, ultimately providing users with a personalized intelligent mobility experience.
Enhanced System Architecture with On-Device AI Capabilities
General large language models serve as the foundation of ECARX AutoGPT's capabilities. ECARX has already established collaborations with foundational AI models such as Microsoft Azure, Llama, Tencent Hunyuan, and Phi. The integration of the DeepSeek-R1 model now allows complex intent recognition and logical reasoning tasks-previously handled via cloud-based models-to run directly on-device.
Given the strict data privacy regulations in different countries and regions, localizing data processing significantly reduces compliance costs for automakers entering overseas markets. This shift decreases reliance on cloud resources, lowers data transmission and storage costs, and provides users with a lower-latency, more secure AI experience.
DeepSeek-R1's Advanced Reinforcement Learning Capabilities
DeepSeek-R1 is extensively trained using reinforcement learning, featuring a well-defined reward mechanism, self-correction, reflection, and deep chain-of-thought (CoT) reasoning. These attributes significantly enhance ECARX AutoGPT's AI scenario engine, particularly in complex cross-domain decision-making. The integration of DeepSeek-R1 dramatically improves AutoScene's scene comprehension, reasoning analysis, and decision recommendation capabilities.
Next-Level AI Interaction & Real-Time User Training
At the product experience level, DeepSeek-R1 has been trained with a "think-first, then-answer" approach. Its chain-of-thought (CoT) reasoning ability enables AutoGPT to perform advanced inference based on diverse user interaction scenarios. Furthermore, it supports the transparency of its reasoning process, allowing users to provide real-time feedback and intervention to improve response accuracy.
This breakthrough represents a major interactive innovation, establishing a highly efficient, real-time two-way feedback mechanism between users and AutoAgent. Users can train their own AI agent on the go, optimizing models and enhancing their personal AI experience dynamically.
Empowering AutoGPT with Specialized Automotive Knowledge
Leveraging DeepSeek-R1's Mixture of Experts (MoE) architecture, AutoAgent's expertise in handling complex in-vehicle interaction tasks will see further improvements. ECARX AutoGPT integrates over 2,000 in-vehicle system interfaces and more than 100,000 pieces of professional knowledge spanning pre-sales and after-sales vehicle support. This dramatically enhances AutoGPT's accuracy and professionalism in automotive mobility scenarios.
A Strategic Leap for ECARX and the Future of Intelligent Mobility
By integrating the DeepSeek-R1 model, ECARX AutoGPT has achieved a substantial leap in inference, decision-making, and vertical expertise, significantly enhancing the user experience. This advancement not only brings cost-efficiency and data privacy advantages to automakers but also aligns with ECARX's global strategic vision of deeply embedding AI technology into its business ecosystem.
This move provides global automakers with smarter, safer, and more flexible in-vehicle AI solutions while fostering an open and collaborative intelligent mobility ecosystem. Through forward-thinking innovation, ECARX is accelerating the evolution of automotive intelligence worldwide.
