理想汽车LIVIS Day：具身AI从座舱向自动驾驶演进

Date16 June 2026 Li Auto Bloomberg2015 HK Reuters2015.HK Livis Day: Advancing Embodied AI fromCockpit to Autonomous Driving Bin WangResearch Analyst+852-220-35496 Livis Day Frames Li Auto’s Embodied AI Vision On June 15, Li Auto held Livis Day, its dedicated software and AI event, toannouncecomprehensive upgrades in software,autonomous driving,andembodied AI. The company’s vision for true embodied AI is rooted in three coreprinciples: safety, capability, and efficiency. For safety, embodied AI must behuman-centric and ultimately safer than humans. In terms of capability, it needsto learn human skills, understand user intent, and independently accomplish tasks Mach Models Extend In-Car AI CapabilitiesLiAuto introduced several key embodied AI models.Its self-developed “MindGPT”has evolved into two models:“Mach Mind-Pro”and“Mach Mind-Edge”.“Mach Mind-Pro”focuses on hearing, language comprehension, memory,and native Agent capabilities, with top-tier performance in general foundationalcapabilitiesand stronger performance than most mainstream models inspecialized Agent benchmarkssuch as Claw series benchmarks and PinchBench-v2 real-world tests. It covershigh-frequency in-car scenarios including vehiclecontrol, smart mobility, working/office, Q&A, and entertainment, while also “Mach VLA”Advances Toward Integrated Autonomous Driving IntelligenceLi Auto’s “MachVision-Language-Action (VLA)” model, launched alongside the all-new“L9”,provides core capabilities in spatial cognition,computational reconstruction, Li Auto has significantly reduced visual input latency, drive-by-wire chassis response latency, OS scheduling and orchestration latency, andmodel inference latency, optimizing overall end-to-end latency by 40%. Thecompany also highlightedmajor capability upgrades, including dual M100 chipswith 2,560 TOPS of compute, 50% larger imitation-learning scale, 15x morereinforcement-learning data, 10x larger parameter count, and 15x higher modelcompute. Architecturally, Li Auto has unified“Mach VLA”,“Mach WorldModel”, “M100”Chip and“Halo OS”Strengthen Full-Stack Control The company also highlighted advancements in chip architecture, operatingsystems, and safety. It developed the“M100”chip, described as the world’s firstdynamicdataflow AI chip,built on a 5nm automotive-grade process anddelivering 1,280 TOPS of compute per chip. When running Li Auto’s proprietaryautonomous driving model, the“M100”achieves performance several timeshigher than“Thor U”, while also supporting general-purpose large models andbroader AI scenarios beyond autonomous driving. Li Auto has also built an end- OTA Roadmap Targets Faster and More Human-Like Driving Li Auto also outlined an ambitious OTA roadmap. In July, autonomous drivingefficiency is expected to improve by 30% compared with the previous version. InSeptember, the system is expected to learn more directly from human behaviorto independently handlecomplex scenarios, including actively avoiding potholesor uneven road sections and enabling full-scenario autonomous reversing, while Appendix 1 Important Disclosures Prices are current as of the end of the previous trading session unless otherwise indicated and are sourced from localexchanges via Reuters, Bloomberg and other vendors. Other information is sourced from Deutsche Bank, subject Important Disclosures Required by U.S. Regulators14-Deutsche Bank and/or its affiliate(s) has received compensation from this company within the past year for non- Important Disclosures Required by Non-U.S. Regulators 24-Deutsche Bank and/or its affiliate(s) is or has been over the previous 12 months party to an agreement with thecompany relating to the provision of services set out in Sections A and B of Annex I of Directive 2014/65/EU, or has For disclosures pertaining to recommendations or estimates made on securities other than the primary subject of thisresearch, please see the most recently published company report or visit our global disclosure look-up page on ourwebsite athttps://research.db.com/Research/Disclosures/EquityResearchDisclosures. Aside from within this report, Analyst Certification The views expressed in this report accurately reflect the personal views of the undersigned lead analyst(s). Inaddition, the undersigned lead analyst(s) has not and will not receive any compensation for providing a specific Company Rating and Dispersion Key The above table provides a snapshot of Deutsche Bank's company research rating distribution across our coveredcompanies. We also present the percentage of companies where Deutsche Bank has provided Investment Banking Note-percentages are rounded so may not total 100%. Covered: The overall rating distribution across all companies under coverage with a rating. w/Banking relation: Percentage of companies under coverage with a rating within each of the "buy", "hold" and "sell"categories for which Deutsche Bank has provided Investment Banking Services within the previous 12 months.