88Z AI, 1 September — Tencent's WeChat Vision Team has open-sourced WeMM-Embedding, available in 2B, 4B, and 9B sizes. It supports unified multimodal encoding and ranks first on the MMEB-v2 benchmark.
At a glance
- WeMM-Embedding supports unified encoding for text, images, and video, ranking first on the MMEB-v2 benchmark.
- Built on Qwen3.5, the model offers 2B/4B/9B sizes and Matryoshka flexible dimension output.
The update
Tencent's WeChat Vision Team has officially open-sourced WeMM-Embedding, a general-purpose multimodal embedding model. Available in 2B, 4B, and 9B configurations, it is designed to meet diverse application needs. The model enables unified encoding for text, images, videos, visual documents, and interleaved multimodal inputs, ensuring efficient cross-modal information processing. In terms of performance, WeMM-Embedding ranks first on the MMEB-v2 benchmark, demonstrating robust multimodal understanding capabilities. Built on the Qwen3.5 architecture, it employs a two-stage training strategy to optimize results. Additionally, it supports Matryoshka flexible dimension output, offering developers greater flexibility in deployment.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- WeMM-EmbeddingSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.