WeChat just pulled back the curtain on its self-developed WeLM large language model family, revealing two configurations. The WeLM-80B, with 80 billion total parameters, activates only 3 billion per inference and is already deployed inside WeChat's native AI agent, Xiaowei. That assistant uses it to power chat search, call WeChat's native features, and connect with mini-program services. The bigger WeLM-617B, still in development, bumps things up to 617 billion total parameters with 23 billion active at a time, using a Mixture-of-Experts architecture. WeChat says the larger model will focus on stronger general understanding and reasoning, and is aimed at more complex WeChat ecosystem tasks — including intelligently developing mini-programs and generating tools for Xiaowei. Tencent also touched on the models during its Q2 earnings report today.