Image-Text, Lite

Cite this benchmark

@inproceedings{xiao2025mieb,
  author = {Xiao, Chenghao and Chung, Isaac and Kerboua, Imene and Stirling, Jamie and Zhang, Xin and Kardos, M\'arton and Solomatin, Roman and Al Moubayed, Noura and Enevoldsen, Kenneth and Muennighoff, Niklas},
  booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
  month = {October},
  pages = {22187-22198},
  title = {MIEB: Massive Image Embedding Benchmark},
  year = {2025},
}

Multilingual image embedding quality across the same task types as MIEB(Multilingual), designed to be run at a fraction of the cost while maintaining relative model rankings.

Reference paper →

Languages 41

Tasks 51

Task Types 10

Models 0