Sentence Embedding Evaluation on MTEB (test)
90.37Classification ScoreNV-Embed-v2
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| NV-Embed-v2embedding dimensions=4096, training data=Non-Fully Public2025.01 | 90.37 | 60.65 | 62.65 | 84.31 | 58.46 | 88.67 | 30.7 | 72.31 | |
| SFR-Embedding-2_Rembedding dimensions=4096, training data=Non-Fully Public2025.01 | 89.05 | 60.14 | 60.18 | 81.26 | 56.17 | 88.07 | 30.71 | 70.31 | |
| BGE-en-ICLembedding dimensions=4096, training data=Fully Public, zero-shot=true2025.01 | 88.62 | 59.66 | 61.67 | 83.74 | 57.51 | 86.93 | 30.75 | 71.24 | |
| LENS-8000embedding dimensions=8000, training data=Fully Public2025.01 | 88.43 | 60.91 | 61.86 | 84.67 | 58.02 | 87.98 | 29.54 | 71.62 | |
| LENS-4000embedding dimensions=4000, training data=Fully Public2025.01 | 88.13 | 60.86 | 60.76 | 84.35 | 57.92 | 87.93 | 31.56 | 71.21 | |
| bge-multilingual-gemma2embedding dimensions=3584, training data=Fully Public2025.01 | 88.08 | 59.72 | 59.24 | 83.88 | 54.65 | 85.84 | 31.2 | 69.88 | |
| stella_en_1.5B_v5embedding dimensions=8192, training data=Non-Fully Public2025.01 | 87.63 | 61.21 | 61.01 | 84.51 | 57.69 | 88.07 | 31.49 | 71.19 | |
| NV-Embed-v1embedding dimensions=4096, training data=Fully Public2025.01 | 87.35 | 60.59 | 59.36 | 82.84 | 52.8 | 86.91 | 31.2 | 69.32 | |
| stella_en_400M_v5embedding dimensions=8192, training data=Non-Fully Public2025.01 | 86.67 | 60.16 | 58.97 | 84.22 | 56.7 | 87.74 | 31.66 | 70.11 | |
| gte-Qwen2-7B-instructembedding dimensions=3584, training data=Non-Fully Public2025.01 | 86.58 | 61.42 | 60.25 | 83.04 | 56.92 | 85.79 | 31.35 | 70.24 | |
| jina-embedTraining Data Composition=synthetic data & public available retrieval data2025.08 | 82.6 | 58.1 | 53.9 | 85.8 | 45.3 | 84 | 29.7 | 65.5 | |
| voyage-large-2-instructembedding dimensions=1024, training data=Non-Fully Public2025.01 | 81.49 | 60.09 | 58.28 | 84.31 | 53.35 | 89.24 | 30.84 | 68.23 | |
| Gecko_lb-768FT. Data Sample Num=>6.6M, Training Data Composition=synthetic data & public available retrieval data2025.08 | 81.2 | 58.9 | 55.7 | 85.1 | 47.5 | 87.6 | 32.6 | 66.3 | |
| Linq-Embed-Mistralembedding dimensions=4096, training data=Non-Fully Public2025.01 | 80.2 | 60.29 | 60.19 | 84.97 | 51.42 | 88.35 | 30.98 | 68.17 | |
| GritLM-7Bembedding dimensions=4096, training data=Fully Public2025.01 | 79.46 | 60.49 | 57.41 | 83.35 | 50.61 | 87.16 | 30.37 | 66.76 | |
| MGHFT. Data Sample Num=820K, Training Data Composition=synthetic data & public available retrieval data2025.08 | 78.8 | 59.8 | 57.5 | 85.6 | 50.1 | 87.9 | 31.3 | 67 | |
| MGHFT. Data Sample Num=310K, Training Data Composition=synthetic data only2025.08 | 78.6 | 60.1 | 51.2 | 82.3 | 49.7 | 86.1 | 31.6 | 64.5 | |
| E5_mistral-7bFT. Data Sample Num=1.8M, Training Data Composition=synthetic data & public available retrieval data2025.08 | 78.5 | 60.2 | 56.9 | 84.6 | 50.3 | 88.3 | 31.4 | 66.6 | |
| E5-mistral-7b-instructembedding dimensions=4096, training data=Non-Fully Public2025.01 | 78.47 | 60.21 | 56.9 | 84.66 | 50.26 | 88.34 | 31.4 | 66.63 | |
| SPEEDFT. Data Sample Num=2.2M, Training Data Composition=synthetic data & public available retrieval data2025.08 | 78.4 | 60.8 | 56.5 | 85.5 | 49.3 | 88.2 | 31.1 | 66.5 | |
| SPEEDFT. Data Sample Num=920K, Training Data Composition=synthetic data only2025.08 | 78.3 | 59.8 | 48.1 | 82.6 | 48.6 | 86.3 | 31.7 | 63.4 | |
| E5_mistral-7bFT. Data Sample Num=500K, Training Data Composition=synthetic data only2025.08 | 78.2 | 59 | 46.9 | 81.2 | 50.5 | 86 | 31.9 | 63.1 | |
| Mistral_gpt-4oFT. Data Sample Num=230K, Training Data Composition=synthetic data only2025.08 | 77.7 | 58.7 | 46.7 | 80.9 | 47.7 | 83.9 | 30.7 | 62.2 | |
| LLM2Vec-Mistral-supervisedembedding dimensions=4096, training data=Fully Public2025.01 | 76.63 | 58.42 | 55.99 | 84.09 | 45.54 | 87.99 | 29.96 | 64.8 | |
| text-embedding-3_largeTraining Data Composition=synthetic data & public available retrieval data2025.08 | 75.5 | 59.2 | 55.4 | 81.7 | 49 | 85.7 | 29.9 | 64.6 | |
| Sentence-T5Supervision=Supervised, Model Size=xxl2022.12 | 73.4 | 56.4 | 42.2 | 82.6 | 43.7 | 85.1 | 30.1 | 59.5 | |
| E5Supervision=Supervised, Model Size=large2022.12 | 73.1 | 56.5 | 50 | 82.1 | 43.3 | 85.9 | 31 | 61.4 | |
| E5Supervision=Supervised, Model Size=base2022.12 | 72.6 | 55.7 | 48.7 | 81 | 42.1 | 85.1 | 31 | 60.4 | |
| Sentence-T5Supervision=Supervised, Model Size=large2022.12 | 72.3 | 54 | 36.7 | 81.8 | 41.7 | 85 | 29.6 | 57.1 | |
| E5Supervision=Supervised, Model Size=small2022.12 | 71.7 | 54.5 | 46 | 80.9 | 39.5 | 85.1 | 31.4 | 58.9 | |
| OmniSONARModel Category=Translation-based2026.03 | 71.143 | — | — | 82.163 | — | 69.037 | — | 74.114 | |
| Gecko_lb-768FT. Data Sample Num=6.6M, Training Data Composition=synthetic data only2025.08 | 70.3 | 57.6 | 53.2 | 83.1 | 46.8 | 86.2 | 32.2 | 62.6 | |
| E5-PTSupervision=Unsupervised, Model Size=large2022.12 | 69 | 54.4 | 44.2 | 69.9 | 44.3 | 80.3 | 32.6 | 56.6 | |
| BERT-FTSupervision=Supervised, Model Size=base2022.12 | 68.7 | 50.5 | 41.5 | 79.2 | 33.9 | 82.6 | 29 | 55.2 | |
| EmbeddingGemmaModel Category=General-purpose2026.03 | 68.108 | — | — | 84.418 | — | 78.644 | — | 77.057 | |
| E5-PTSupervision=Unsupervised, Model Size=base2022.12 | 67.9 | 53.5 | 42.9 | 69.5 | 43.4 | 79.2 | 31.1 | 55.6 | |
| DALRBackbone=BERTbase, Training Data=wiki+flickr2025.06 | 67.46 | 48.35 | 23.84 | 79.38 | — | — | — | — | |
| GTR_xxlFT. Data Sample Num=662K, Training Data Composition=synthetic data & public available retrieval data2025.08 | 67.4 | 56.7 | 48.5 | 78.4 | 42.4 | 86.1 | 30.6 | 59 | |
| GTRSupervision=Supervised, Model Size=xxl2022.12 | 67.4 | 56.7 | 48.5 | 78.4 | 42.4 | 86.1 | 30.6 | 59 | |
| SimCSE-BERT-supSupervision=Supervised2022.12 | 67.3 | 47.5 | 21.8 | 79.1 | 33.4 | 73.7 | 23.3 | 48.7 | |
| GTRSupervision=Supervised, Model Size=large2022.12 | 67.1 | 55.4 | 47.4 | 78.2 | 41.6 | 85.3 | 29.5 | 58.3 | |
| E5-PTSupervision=Unsupervised, Model Size=small2022.12 | 67 | 53.1 | 40.8 | 68.8 | 41.7 | 78.2 | 32.7 | 54.3 | |
| ContrieverSupervision=Supervised2022.12 | 66.7 | 53.1 | 41.9 | 76.5 | 41.1 | 82.5 | 30.4 | 56 | |
| KDMCSE w/ IMLBackbone=BERTbase, Training Data=wiki+flickr2025.06 | 65.32 | 47.96 | 22.67 | 78.81 | — | — | — | — | |
| MEXMAModel Category=Translation-based2026.03 | 65.197 | — | — | 75.678 | — | 63.078 | — | 67.984 | |
| KDMCSEBackbone=BERTbase, Training Data=wiki+flickr2025.06 | 64.83 | 47.5 | 22.06 | 78.34 | — | — | — | — | |
| Qwen3-Embedding-0.6BModel Category=General-purpose2026.03 | 64.707 | — | — | 84.724 | — | 76.358 | — | 75.263 | |
| mE5largeModel Category=General-purpose2026.03 | 63.672 | — | — | 82.502 | — | 73.527 | — | 73.234 | |
| MCSEBackbone=BERTbase, Training Data=wiki+flickr2025.06 | 63.2 | 46.92 | 21.43 | 77.02 | — | — | — | — | |
| SimCSEBackbone=BERTbase, Training Data=wiki+flickr2025.06 | 62.54 | 46.47 | 20.29 | 74.33 | — | — | — | — | |
| SimCSE-BERT-unsupSupervision=Unsupervised2022.12 | 62.5 | 46.5 | 20.3 | 74.3 | 29 | 70.3 | 31.2 | 45.5 | |
| BERTSupervision=Unsupervised2022.12 | 61.7 | 43.4 | 10.6 | 54.4 | 30.1 | 56.3 | 29.8 | 38.3 | |
| LaBSEModel Category=Translation-based2026.03 | 61.502 | — | — | 74.182 | — | 64.662 | — | 66.782 | |
| SONARModel Category=Translation-based2026.03 | 58.479 | — | — | 70.824 | — | 60.712 | — | 63.338 | |
| GloveSupervision=Unsupervised2022.12 | 57.3 | 43.3 | 21.6 | 61.9 | 27.7 | 70.9 | 28.9 | 42 | |
| HS (Full)Backbone=Llama-22026.02 | — | — | — | — | — | — | — | 31.25 | |
| MetaEOLBackbone=Llama-22026.02 | — | — | — | — | — | — | — | 50.83 | |
| MetaEOLBackbone=Qwen-32026.02 | — | — | — | — | — | — | — | 52.72 | |
| VABackbone=Llama-2, Aggregation Strategy=Layer Selection2026.02 | — | — | — | — | — | — | — | 52.25 | |
| VABackbone=Qwen-3, Aggregation Strategy=Layer Selection2026.02 | — | — | — | — | — | — | — | 51.59 |