vivo's Self-Developed AI Large Model to be First Applied in OriginOS 4 System


In the first three quarters of 2025, China's smart speaker sales reached 10.54 million units, and the annual sales may reach 14.2 million. However, the industry faces a critical challenge: only 33% of devices are equipped with AI large models, and nearly 70% still rely on basic voice interaction, indicating insufficient popularization of intelligence. The high-end market was driven by "Super Xiaoai," with large models becoming a new selling point.
At the 2025 vivo Developer Conference, Vice President Zhou Wei announced significant breakthroughs in edge AI for smartphones, emphasizing the trend of shifting from cloud-based large models to lightweight edge models. This move aims to enhance user experience and drive differentiated competition. vivo launched the Blue Heart Smart Strategy and the OriginOS6 system, with edge AI models being the core highlight.
At the 2025 vivo Developer Conference, the Blue Heart Language Large Model was upgraded, and the smart assistant XIAO V achieved significant progress. Core improvements include restructuring the intent control center, enhancing the accuracy of user intent understanding, enabling the decomposition of complex tasks and optimization of execution steps. A new deep thinking ability has been added, allowing XIAO V to provide more insightful and high-quality intelligent Q&A services.
vivo launched the Blue Heart 3B On-Device Multimodal Reasoning Large Model at the 2025 Developer Conference. This 3 billion parameter model is the first One Model in the industry to integrate five core capabilities. After one year of training and optimization, it achieves a major breakthrough in deploying complex multimodal AI capabilities on mobile devices, establishing a leading position in the industry.
At the 22nd China Cybersecurity Annual Conference (and the Cybersecurity Coordinated Defense Sub-forum of the National Cybersecurity Publicity Week), the first crowdsourced testing results for AI large models in China were announced. This event was guided by the Cybersecurity Coordination Bureau of the Central Cyberspace Administration Office and hosted by the National Computer Network Emergency Response Technology Coordination Center. It attracted 559 white hat security experts who conducted comprehensive security vulnerability tests on 15 AI large models and application products. The testing covered a variety of products, ranging from basic large models to intelligent agents and model development platforms, aiming to examine them from an attacker's perspective.