Today, at the vivo Developer Conference, the spotlight shone on the official launch of the BlueLM 3B on-device multimodal reasoning large model. Following a year of relentless training and fine-tuning, this model has emerged as the industry's pioneer in achieving a seamless integration of five essential capabilities—language comprehension, multimodal processing, logical reasoning, 128K long-context understanding, and UI Agent functionality—all within a compact 3-billion-parameter framework. On the esteemed multimodal benchmark OpenCompass, the BlueLM 3B has outperformed all models with 8 billion parameters, establishing itself as the undisputed leader among large models with fewer than 10 billion parameters. Moreover, when evaluated on the SuperCLUE benchmark, which focuses on on-device large models for smartphones, the BlueLM 3B claimed the top spot on the overall leaderboard for models with fewer than 10 billion parameters. In addition to these remarkable achievements, vivo's AI has been honored with the highest rating of 'L3 Excellence Level' in the evaluation conducted by the China Academy of Information and Communications Technology (CAICT) for 'Terminal Intelligent Service Capability Grading.'