WeChat Open-Sources WeMM-Embedding Model, Handling 1 Billion Calls Daily

WeChat's AI team announced on September 4 that it has open-sourced WeMM-Embedding, a multimodal embedding model with 2B, 4B, and 9B versions. It handles text, images, videos, visual documents, and mixed inputs, mapping them into one semantic space for easier comparison and search.

The model is already used widely in WeChat, powering features like Moments search, video recommendations, official account suggestions, and e-commerce. It handles over 1 billion calls daily. On the MMEB-v2 benchmark, it ranks first, beating previous open and closed models. The research paper, code, and model weights are now public for the research community.