晚点独家丨理想调整基座模型业务：詹锟接手，VLA 研发整合

Core Insights - The article discusses the strategic shift at Li Auto, focusing on the integration of the VLA (Vision-Language-Action) model into its AI strategy, which aims to enhance the company's capabilities in autonomous driving and intelligent cockpit technologies [4][6]. Group 1: Leadership Changes and Organizational Structure - Jian Kun, a senior algorithm expert in autonomous driving, will take over the VLA model business, reporting directly to the CTO, Xie Yan, instead of the previous head of autonomous driving, Lang Xianpeng [4]. - The VLA model development was previously divided between two departments, with the language component led by Chen Wei and the vision and action components managed by the autonomous driving team [6]. - The VLA model has been elevated from a single project to a company-wide foundational model, indicating a clearer focus on embodied intelligence as a core part of Li Auto's AI strategy [6]. Group 2: VLA Model and Future Developments - The VLA model is designed to support embodied intelligence, with "V" representing perception, "L" for planning and decision-making, and "A" for action, applicable to both physical and digital realms [6]. - Li Auto plans to fully deploy the VLA driver model in its AD Max vehicles by September 2025, aiming for smoother longitudinal control and improved accuracy at complex intersections [8]. - The company intends to enhance language and behavioral interactions and adapt its self-developed M100 chip by 2026 [8]. Group 3: Talent Movement and Industry Context - Several technical talents have left Li Auto following the AI model boom, but some maintain close communication with the company and participate in industry sharing sessions [7]. - Chen Wei, the outgoing head of the VLA model, previously emphasized the importance of building a complete agent system centered around the VLA model, viewing all products as robotic entities [7].