WeMM Embedding

WeMM Embedding is a family of universal multimodal embedding models by Tencent providing unified representations for text, images, videos, and visual documents.

WeMM-Embedding is a family of universal multimodal embedding models developed by the WeChat Vision team. It provides unified representations for text, images, videos, visual documents, and interleaved multimodal inputs. The models achieve high performance across diverse tasks and benchmarks, supporting Matryoshka dimensions for flexible truncation. Embeddings are extracted from the last-layer hidden state at a dedicated token position and normalized. The models are available in 2B, 4B, and 9B parameter sizes and integrate with Hugging Face transformers, Sentence Transformers, vLLM, and SGLang. Audio input is not currently supported.

WeMM Embedding at a glance

Best for: Software engineer, Researcher
Runs on: API, self-hosted, open source
Pricing: Free
Released: August 2026
Category: AI research tool
Website: github.com/tencent/wemm-embedding

Key features

Supports text, image, video, and visual document inputs
Provides Matryoshka representation learning with flexible dimensions
Integrates with Hugging Face transformers and Sentence Transformers
Supports serving via vLLM and SGLang
Generates unified multimodal embeddings via L2 normalization

Use it for

Multimodal retrieval across diverse tasks and domains
Generating text, image, and video embeddings
Evaluating model performance on MMEB-v2 and MMEB-v3 benchmarks

Alternatives to WeMM Embedding

MTEB — Evaluates embedding models across over 1000 tasks and 1000 languages.
Vlmhub — Run multimodal prompts across multiple local and cloud backends without rewriting inference code.
Hugging Face — Home of open-source AI models and the Transformers library
WeKnora — Combines RAG, multi-step agents, and automatic Wiki generation in one platform.
VLMEvalKit — Evaluate large vision-language models across numerous benchmarks with a single command.
OpenVLHarness — Lifts six multimodal model backbones by 9.7 to 12.6 points across 22 benchmarks.
Every number in a tiny GPT — Inspects every lookup, matrix multiplication, and normalisation in a live transformer.
Overview — Learn language models by running hands-on experiments without writing scripts
LMMs-Eval — Provides open research and tooling for multimodal model evaluation and benchmarks.
Ncm — Generates text significantly faster than diffusion baselines while maintaining high quality.
Anime Text Dataset — Provides structured anime data with field-level provenance and question-answering pairs.
RAG-Anything — It unifies text and multimodal document retrieval within a single framework.

All WeMM Embedding alternatives