MMEB Leaderboard: leaderboard

Massive Multimodal Embedding Benchmark leaderboard evaluating image, video, and visual-document embedding models across classification, retrieval, QA, grounding, and document tasks.

Metric: Overall Score (%). Source: huggingface.co. 95 models tracked.

Top models

#ModelScore
1WeMM-Embedding-9B59.55
2Ovis-Omni-Embedding-3B58.46
3WeMM-Embedding-4B58.21
4Ovis-Omni-Embedding-v0.5-3B57.42
5WeMM-Embedding-2B56.02
6Tianmu-Emb-Uni53.27
7Ovis-Omni-Embedding-v0.1-3B53.09
8UEmbed-9B (hybrid)52.26
9UEmbed-9B (dense)52.22
10UEmbed-9B (sparse)51.77

Interactive version: theaggregate.ai/benchmark?slug=mmeb-leaderboard · How It Works · Data refreshed daily, snapshot 2026-09-05.