Tencent WeChat Team Releases WeLM LLM Focusing on Resource Efficiency

Tencent's WeChat team recently unveiled its self-developed large language model, WeLM, on platform X, emphasizing resource efficiency in its design. Two configurations were disclosed: WeLM-80B with 80 billion total parameters and 3 billion active parameters, and WeLM-617B with 617 billion total parameters and 23 billion active parameters. The WeLM-80B is already integrated into WeChat's native AI assistant, Xiaowei, supporting chat, search, and access to WeChat features and mini-programs. The WeLM-617B is under development, using a mixture-of-experts architecture to boost reasoning and understanding while keeping active parameters moderate, targeting complex tasks within the WeChat ecosystem, such as smart mini-program development and Xiaowei tool generation. This move reflects Tencent's ongoing investment in AI to optimize efficiency and user experience.