vivo Previews BlueCode Mobile AI Agent and 30B MoE Smartphone LLM
vivo previewed BlueCode, a native AI programming agent designed to bring software development tools directly to smartphones. The company also announced ongoing research into running a 30-billion-parameter Mixture of Experts (MoE) BlueLM model on mobile hardware. Deploying a 30B-parameter model on a mobile device pushes the boundaries of on-device AI performance, allowing complex code generation without cloud dependency. This highlights the effort by smartphone makers to transform mobile devices into independent productivity hubs while improving latency and privacy. The announcement was made alongside updates to vivo's BlueLM strategic roadmap, OriginOS 7, and the self-developed BlueOS 4. Using a Mixture of Experts architecture allows the model to activate only a subset of parameters during inference, making large models far more practical under smartphone compute and RAM constraints.
## BACKGROUND
Mixture of Experts (MoE) is a neural network architecture that divides a large model into specialized sub-networks called experts, routing input tokens only to the relevant experts to reduce active computational cost. vivo's BlueLM is a proprietary family of large language models created by vivo AI Lab for general intelligence tasks. On-device AI processing minimizes latency, enables offline operation, and preserves privacy by keeping user data local.