Embedding benchmarks
Every benchmark the Embedding index tracks, with each core benchmark's share of the composite and how many models it covers.
33 benchmarks tracked.
- BEIR nDCG@10 mean
- BRIGHT(v1.1) nDCG@10 mean
- CoIR nDCG@10 mean
- JinaVDR nDCG@5 mean
- LongEmbed mean
- MIEB(lite) task mean
- MIEB(Multilingual) task mean
- MMEB-V2 Overall
- MMEB-V2 Image
- MMEB-V2 Video
- MMEB-V2 VisDoc
- MTEB(cmn, v1) (C-MTEB successor) task mean
- MTEB(Code, v1) nDCG@10 mean
- MTEB(eng, v1) task mean
- MTEB(eng, v2) task mean
- MTEB(eng, v2) Classification
- MTEB(eng, v2) Clustering
- MTEB(eng, v2) Pair classification
- MTEB(eng, v2) Retrieval
- MTEB(eng, v2) STS
- MTEB(Europe, v1) task mean
- MTEB(Multilingual, v1) task mean
- MTEB(Multilingual, v2) task mean
- MTEB(Multilingual, v2) Bitext mining
- MTEB(Multilingual, v2) Classification
- MTEB(Multilingual, v2) Clustering
- MTEB(Multilingual, v2) Retrieval
- MTEB(Multilingual, v2) STS
- RTEB(beta) nDCG@10 mean
- RTEB(eng, beta) nDCG@10 mean
- ViDoRe v1+v2 nDCG@5 mean
- ViDoRe V3 nDCG@10 mean (10 public tasks)
- ViDoRe v3.1 nDCG@10 mean