At the Software and Embodied Intelligence Conference, Li Auto announced that it will drive the continuous evolution of Mach VLA capabilities in the second half of the year. Specifically, the third quarter will see the rollout of a new Mach VLA for AD Max, and by the fourth quarter, it will achieve alignment with Tesla's FSD V14 capabilities. The dual Mach M100 chips deliver powerful computing performance, reaching 2560 TOPS, with a 50% increase in imitation learning scale, a 15-fold increase in reinforcement learning scale, and a 10-fold and 15-fold increase in model parameters and computational volume, respectively. Meanwhile, Li Auto unveiled its new self-developed foundational large model—the Mach Mind-4 series, including Mach Mind-Pro and Mach Mind-Edge. Mach Mind-Pro consistently ranks among the top tier in the industry across multiple authoritative benchmark tests, outperforming most mainstream Agent models in terms of overall performance and leading in four core business dimensions. Its native Agent intelligence capabilities have been fully implemented, empowering the L9 Livis in-vehicle intelligence. Mach Mind-Edge, on the other hand, is the industry's first truly mass-produced end-side native embodied intelligent agent, utilizing multi-modal streaming temporal modeling. It possesses the ability to continuously understand the dynamic physical world, perform causal reasoning, and make autonomous decisions, directly outputting actions and real-time vehicle hardware control. The model enables Always-On 24/7 proactive perception, human-vehicle interaction, autonomous vehicle control, and multi-modal Q&A, with all functions completed locally on the vehicle side without uploading data to the cloud. It is not a scaled-down version of a cloud model but is instead natively designed for in-vehicle scenarios from the ground up.