Weixin introduces the WeMM-Embedding multimodal model
Weixin says its 9B model, already used in several of its services, supports search and recommendations across text, images, and video and ranked first in two MMEB benchmark versions.
Weixin’s Vision team developed WeMM-Embedding, a multimodal embedding model intended for search and recommendations across text, images, and video, according to the company’s announcement.
The model has 9 billion parameters and is deployed in several Weixin services. The company also says it ranked first on the MMEB-v2 and MMEB-v3 benchmark versions.
To assess whether the result applies to your own system, define representative tasks and compare results with a baseline. Also check data quality, latency, and cost; a benchmark ranking does not guarantee the same performance in other settings.
If you use AI to study or apply the material, avoid sending personal data or internal information without authorization. Rota Nacional applies personal-data detection and the organization’s policy before execution; embeddings are in preview, so confirm availability before planning a use.