At the 2026 vivo Developer Conference, vivo proudly presented four cutting-edge BlueLM large models. Notably, the BlueLM speech large model, named RealTime, employs an end-to-end speech comprehension framework, empowering it to adeptly manage intricate speech scenarios. The BlueLM on-device large model, Nano, is distinguished by its 3 billion parameters and multimodal real-time perception abilities, enabling it to amass personal memories. Its associated functionalities are among the top-tier in the industry. The BlueLM cloud-based large models, Flash and Pro, are fortified with a self-developed Agentic engine, facilitating seamless cross-device and cross-application task execution. Moreover, vivo introduced the system-level BlueLM Harness, which now encompasses over 6,000 atomic skills, establishing a system-level perception and personal memory repository for users and propelling the transformation of OriginOS into a personalized intelligent system. Concurrently, vivo is also delving into preliminary research on the BlueLM 30B MoE on-device large model.