ChainThink reports that on August 13, according to the WeChat team, WeChat has unveiled two configurations of its self-developed WeLM large model: the WeLM-80B, with a total of 80 billion parameters and 3 billion parameters activated per inference, has been deployed to WeChat’s native AI agent, “Xiao Wei.”
WeChat Mini Program is currently used for chat search, accessing native WeChat functions, and providing Mini Program services. In its Q2 financial report, Tencent mentioned that WeChat Mini Program is undergoing limited gray testing within WeChat, powered by WeLM, a model customized for user privacy, WeChat-specific scenarios, and inference efficiency.
A larger-scale WeLM-617B is currently under development, featuring a total of 617 billion parameters and 23 billion activated parameters per inference, built on a MoE architecture, primarily designed for more complex WeChat ecosystem tasks, including intelligent mini-program development and tool generation for small businesses.
WeLM-80B and WeLM-617B were previously mentioned in the WeChat team's July Hidden Decoding paper; this update further clarifies the product positioning of both models.