Information Retrieval on SciFact BEIR (NDCG@10)
99.36NDCG@10DIVE
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DIVEDimension=512d, Backbone=LLM2Vec-Mistral-7B2026.05 | 99.36 | — | — | |
| SearchDimension=512d, Backbone=LLM2Vec-Mistral-7B2026.05 | 98.59 | — | — | |
| SMECDimension=512d, Backbone=LLM2Vec-Mistral-7B2026.05 | 92.97 | — | — | |
| inf-v1Model Size Category=<1B parameters2025.10 | 85.4 | — | — | |
| inf-v1-1.5bModel Size Category=>=1B parameters2025.10 | 82.8 | — | — | |
| MatryoshkaDimension=512d, Backbone=LLM2Vec-Mistral-7B2026.05 | 80.05 | — | — | |
| bge-en-iclModel Size Category=>=1B parameters2025.10 | 79.1 | — | — | |
| Qwen3-8BModel Size Category=>=1B parameters2025.10 | 78.5 | — | — | |
| LENS-d4KModel Size Category=>=1B parameters2025.10 | 78.4 | — | — | |
| Qwen3-4BModel Size Category=>=1B parameters2025.10 | 78.3 | — | — | |
| gte-base-v1.5Model Size Category=<1B parameters2025.10 | 76.8 | — | — | |
| gte-baseModel Size Category=<1B parameters2025.10 | 76.2 | — | — | |
| bge-lg-v1.5Model Size Category=<1B parameters2025.10 | 74.6 | — | — | |
| gte-largeModel Size Category=>=1B parameters2025.10 | 74.3 | — | — | |
| MUGIBackbone=GPT-4.12026.06 | 73.5 | — | — | |
| STORMBackbone=Qwen3-8B, Max tokens=642026.06 | 72.8 | — | — | |
| GenQRBackbone=GPT-4.12026.06 | 72.6 | — | — | |
| opensearch-gteModel Size Category=<1B parameters2025.10 | 72.5 | — | — | |
| GenQREBackbone=GPT-4.12026.06 | 72.5 | — | — | |
| LameRBackbone=GPT-4.12026.06 | 72.5 | — | — | |
| e5-base-v2 + SOFTCENTROIDModel=e5-base-v2, Program Strategy=SOFTCENTROID (default), K=3, α=0.5, τ=0.05, Cost=1×, Per-task tuning=No2026.05 | 72.41 | — | — | |
| bge-largeModel Size Category=<1B parameters2025.10 | 72.4 | — | — | |
| CSQEBackbone=GPT-4.12026.06 | 72.1 | — | — | |
| e5-base-v2Model=e5-base-v22026.05 | 72.04 | — | — | |
| Q2DBackbone=GPT-4.1, Evaluation Protocol=Zero-shot2026.06 | 72 | — | — | |
| e5-large-instModel Size Category=>=1B parameters2025.10 | 71.6 | — | — | |
| Splade-V3Model Size Category=<1B parameters2025.10 | 71 | — | — | |
| Q2KBackbone=GPT-4.12026.06 | 70.9 | — | — | |
| QA-ExpaBackbone=GPT-4.12026.06 | 70.6 | — | — | |
| e5-largeModel Size Category=<1B parameters2025.10 | 70.2 | — | — | |
| MILCOModel Size Category=<1B parameters2025.10 | 70.1 | — | — | |
| Qwen3-0.6BModel Size Category=<1B parameters2025.10 | 69.7 | — | — | |
| LLM2VecDimension=4096d, Backbone=LLM2Vec-Mistral-7B2026.05 | 68.9 | — | — | |
| RM3Backbone=GPT-4.12026.06 | 64.6 | — | — | |
| rawEncoder=e5-base, Kcands=2002026.06 | 63.7 | — | — | |
| rawEncoder=e5-small, Kcands=2002026.06 | 59.8 | — | — | |
| SHARDEncoder=e5-base, config=d/4, Kcands=2002026.06 | — | 0.001 | 0 | |
| SHARDEncoder=e5-base, config=d/8, Kcands=2002026.06 | — | 0.007 | 0.017 | |
| SHARDEncoder=e5-small, config=d/4, Kcands=2002026.06 | — | 0.008 | 0.02 | |
| SHARDEncoder=e5-small, config=d/8, Kcands=2002026.06 | — | 0.025 | 0.042 | |
| SVDEncoder=e5-base, k=d/2, Kcands=2002026.06 | — | 0.052 | 0.073 | |
| SVDEncoder=e5-small, k=d/2, Kcands=2002026.06 | — | 0.076 | 0.1 |