vivo previews BlueCode phone coding agent, explores 30B MoE model for on-device use

vivo has previewed BlueCode, a coding agent designed to run on phones, and says it is exploring a roughly 30B-parameter mixture-of-experts model for on-device use, according to TechNode. The report places the two items in the same announcement, pointing to a handset maker working on both a coding-focused agent and a locally running large model.

According to the source, BlueCode is described as a phone-side coding agent, meaning code-related tasks would be handled by an agent running on the device itself, while the 30B MoE model is framed as something under exploration for on-device deployment. The wording used is "preview" and "explores", which indicates neither is a shipping or generally available product. The report does not give a release date, a list of supported devices, or any concrete capability boundaries.

The key signal is the combination of on-device compute and an agent: moving a high-value, high-frequency task like coding onto the phone touches response latency, privacy and offline availability at once. Coding also has an advantage as an early target because correctness is relatively easy for users to judge, which makes product feedback cleaner than in more subjective assistant tasks.

The MoE choice is unsurprising in this context. Because mixture-of-experts models activate only part of their parameters per inference, they can in principle retain more capability while lowering the compute cost of each run, which is exactly what on-device deployment lacks. But a roughly 30B total footprint remains a hard constraint for phone storage and memory bandwidth, so quantization, expert routing and memory management will decide real-world usability, and those are precisely the details the report leaves out.

At the industry level, handset makers have spent the past two years treating on-device models as a differentiator, since cloud capabilities have converged and locally reliable agents are easier for users to feel. vivo choosing a coding scenario suggests vendors are hunting for on-device use cases with more to do than chat or summarization.

Three things are worth watching next: whether BlueCode ships with a system update or a new device, what parameter count and quantization scheme the MoE model ultimately lands on, and whether vivo publishes measured results for on-device coding tasks. Until then, the item is better read as a directional preview than a product launch.