Semantic Textual Similarity on BIOSSES
89.6Spearman CorrelationBioHiCL-base
Evaluation Results
| Method | Links | |
|---|---|---|
| BioHiCL-baseSize=0.1B2026.04 | 89.6 | |
| BiCA-baseSize=0.1B2026.04 | 88 | |
| BiCA-smallSize=0.03B2026.04 | 87.2 | |
| BioLORDType=Black-Box2026.03 | 87.18 | |
| MedEmbedType=Black-Box2026.03 | 86.99 | |
| BioHiCL-largeSize=0.3B2026.04 | 86.8 | |
| bge-base-en-v1.5Size=0.1B2026.04 | 86 | |
| BMRetriever-1BSize=1B2026.04 | 85.8 | |
| bge-large-en-v1.5Size=0.3B2026.04 | 84.4 | |
| BMRetriever-410MSize=0.4B2026.04 | 84 | |
| PubMedBERTType=Black-Box2026.03 | 83.96 | |
| MedCPT-QuerySize=0.1B2026.04 | 83.7 | |
| ContrieverSize=0.1B2026.04 | 83.3 | |
| e5-large-v2Size=0.3B2026.04 | 83 | |
| SGPT-1.3BSize=1.3B2026.04 | 83 | |
| SpladeV3Size=0.3B2026.04 | 82.7 | |
| SimCSE + MixSPEncoder Architecture=BERT-Base, Base Method=SimCSE, Training Strategy=MixSP2024.06 | 82.61 | |
| SapBERTType=Black-Box2026.03 | 82.48 | |
| MedCPTType=Black-Box2026.03 | 81.95 | |
| DiffAug + MixSPEncoder Architecture=BERT-Base, Base Method=DiffAug, Training Strategy=MixSP2024.06 | 81.23 | |
| SimCSE + MixSPEncoder Architecture=RoBERTa-Base, Base Method=SimCSE, Training Strategy=MixSP2024.06 | 80.74 | |
| SBERT + MixSPEncoder Architecture=BERT-Base, Base Method=SBERT, Training Strategy=MixSP2024.06 | 80.58 | |
| EmbeddingGemmaType=Black-Box2026.03 | 80.46 | |
| DiffAug + MixSPEncoder Architecture=RoBERTa-Base, Base Method=DiffAug, Training Strategy=MixSP2024.06 | 80.35 | |
| DiffAug + MoEEncoder Architecture=BERT-Base, Base Method=DiffAug, Training Strategy=MoE2024.06 | 79.95 | |
| QIME-TF-MMRType=Interpretable2026.03 | 79.66 | |
| LDIR-500Type=Interpretable2026.03 | 79.3 | |
| BiomedBERTSize=0.1B2026.04 | 79.1 | |
| gtr-t5-xlSize=1.2B2026.04 | 78.9 | |
| SBERT + MoEEncoder Architecture=BERT-Base, Base Method=SBERT, Training Strategy=MoE2024.06 | 78.52 | |
| SimCSE + MoEEncoder Architecture=BERT-Base, Base Method=SimCSE, Training Strategy=MoE2024.06 | 77.07 | |
| SimCSE + FTEncoder Architecture=BERT-Base, Base Method=SimCSE, Training Strategy=FT2024.06 | 76.63 | |
| SBERT + MixSPEncoder Architecture=RoBERTa-Base, Base Method=SBERT, Training Strategy=MixSP2024.06 | 76.01 | |
| QIME-TFType=Interpretable2026.03 | 75.6 | |
| SBERT + MoEEncoder Architecture=RoBERTa-Base, Base Method=SBERT, Training Strategy=MoE2024.06 | 73.75 | |
| SimCSE + FTEncoder Architecture=RoBERTa-Base, Base Method=SimCSE, Training Strategy=FT2024.06 | 73.35 | |
| SimCSE + MoEEncoder Architecture=RoBERTa-Base, Base Method=SimCSE, Training Strategy=MoE2024.06 | 72.65 | |
| DiffAug + MoEEncoder Architecture=RoBERTa-Base, Base Method=DiffAug, Training Strategy=MoE2024.06 | 72.65 | |
| SBERTEncoder Architecture=RoBERTa-Base, Base Method=SBERT, Training Strategy=Base2024.06 | 72.03 | |
| DiffAug + FTEncoder Architecture=BERT-Base, Base Method=DiffAug, Training Strategy=FT2024.06 | 71.26 | |
| DiffAug + FTEncoder Architecture=RoBERTa-Base, Base Method=DiffAug, Training Strategy=FT2024.06 | 71.02 | |
| SimCSE (Unsup)Type=Black-Box2026.03 | 68.86 | |
| BMRetrieverType=Black-Box2026.03 | 68.85 | |
| Bag-of-WordsType=Interpretable2026.03 | 68.78 | |
| SimCSEEncoder Architecture=BERT-Base, Base Method=SimCSE, Training Strategy=Base2024.06 | 68.38 | |
| SimCSEEncoder Architecture=RoBERTa-Base, Base Method=SimCSE, Training Strategy=Base2024.06 | 67.75 | |
| SimCSE (Sup)Type=Black-Box2026.03 | 67.19 | |
| SBERTEncoder Architecture=BERT-Base, Base Method=SBERT, Training Strategy=Base2024.06 | 63.88 | |
| QIMEType=Interpretable2026.03 | 61.88 | |
| CQG-MBQAType=Interpretable2026.03 | 54.97 | |
| BERTType=Black-Box2026.03 | 54.7 | |
| QA-EmbType=Interpretable2026.03 | 46.43 | |
| GloVeType=Black-Box2026.03 | 44.93 | |
| DiffAugEncoder Architecture=BERT-Base, Base Method=DiffAug, Training Strategy=Base2024.06 | 40.12 | |
| DiffAugEncoder Architecture=RoBERTa-Base, Base Method=DiffAug, Training Strategy=Base2024.06 | 39.15 |