StepFun and Qianli Tech Unveil Full-Stack Large Model Solution for Automotive Applications

On July 13, StepFun—a company founded in 2023—launched three core products tailored for end-device scenarios: the on-device model family Step Edge, the agent-native operating system Step AOS, and the personal AI agent Amoo. On the same day, its strategic partner Qianli Technology announced plans to integrate this full-stack capability into the automotive sector, jointly advancing the deployment of large-model-native architectures in vehicles. Step Edge is an on-device model specifically customized for terminal hardware such as smartphones and cars. It enables lightweight tasks to be processed locally while supporting complex tasks through coordinated edge-cloud execution, delivering low-latency responses in high-frequency in-vehicle scenarios like voice interaction, navigation queries, and vehicle status Q&A. According to StepFun, the model ranks first among comparable on-device models across 29 core evaluation metrics, spanning text understanding, visual recognition, audio comprehension, speech recognition, GUI interaction, image generation, and editing. Step AOS is positioned as “the world’s first agent-native operating system,” offering three core capabilities: memory, decision-making execution, and security. It encapsulates terminal hardware compute resources, sensor data, and software functionalities into callable interfaces for AI agents, allowing users to accomplish tasks autonomously by simply expressing their intent. Amoo, a personal AI agent running atop Step AOS, possesses OS-level permissions, enabling cross-application resource orchestration and seamless edge-cloud collaboration. When deployed in vehicles, Amoo will focus on mobility scenarios, covering the entire user journey—from trip planning and navigation queries to vehicle management. Qianli Technology brings full-stack mass-production capabilities in intelligent driving and smart cockpit domains, including in-vehicle system integration, automotive-grade functional safety certification, and whole-vehicle integration—providing engineering-level implementation support for StepFun’s technology stack. Currently, most in-vehicle AI agents in the industry are limited to application-layer permissions and cannot access underlying systems like autonomous driving or vehicle control modules. This partnership aims to overcome that bottleneck at the system level: Qianli Technology will handle on-device model inference optimization on automotive-grade chips, functional safety certification, and compliant processing of multimodal data. StepFun emphasizes that AI agents must be capable of “getting things done,” not just “chatting.” On mobile devices, it has already established deep collaborations with Ctrip, Alipay, DiDi, Meituan, and WPS, executing cross-scenario tasks via service APIs rather than GUI simulation. Extending this approach to vehicles, the agent can synthesize user preferences, vehicle range, and other contextual information to orchestrate navigation, hotel booking, EV charging station searches, and payment services into a closed-loop experience. On the security front, all operations run within a Trusted Execution Environment (TEE), employing end-to-end encryption, immediate permission revocation, fully traceable and reversible actions, and requiring additional user authorization for high-risk operations—with ultimate decision-making authority always retained by the user. This collaboration marks a new model of cooperation between AI companies and Tier 1 suppliers: the former provides model and system-layer capabilities, while the latter handles engineering implementation and automotive-grade adaptation, jointly defining the technical architecture for next-generation intelligent vehicles.