Tencent's wechat team releases open-source multimodal model with high daily usage

On September 7, 2026, Tencent's WeChat Vision Team announced the open-source release of the general-purpose multimodal embedding model WeMM-Embedding. The model supports text and image inputs, aiming to support AI developers with advanced architecture and training strategies. WeMM-Embedding is available in three versions: 2B, 4B, and 9B, designed for varied deployment needs. The model uses two-stage training and knowledge distillation to enhance understanding while balancing efficiency. The model, which has been tested in real business scenarios, is already widely used in WeChat's recommendation and search systems, with daily call volume exceeding ten billion. The open-source release highlights WeChat's commitment to building a multimodal AI ecosystem and offers technical support for industry AI development.