Multimodal Retrieval on MMEB Image V2
78.2Overall ScoreQwen3-VL-Reranker-8B
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen3-VL-Reranker-8BModel Scale=8B2026.06 | 78.2 | — | — | — | — | |
| Qwen3-VL-8B-rerankerModel Scale=8B, Fine-tuning Recipe=our training recipe2026.06 | 76.2 | 70 | 77.8 | 75.2 | 90.8 | |
| Qwen3-VL-4B-rerankerModel Scale=4B, Fine-tuning Recipe=our training recipe2026.06 | 75.4 | 69.7 | 76.1 | 75.2 | 88.7 | |
| Qwen3-VL-8B-miniRerankerModel Scale=8B, Fine-tuning Recipe=our training recipe, Reranker Type=miniReranker2026.06 | 74.3 | 69.2 | 73.9 | 73.7 | 89.7 | |
| UME-R1Backbone=Qwen2-VL, Parameter Scale=7B, Oracle Evaluation Mode=true2025.11 | 74.2 | 69.1 | 73.2 | 74.8 | 87.4 | |
| Qwen3-VL-Reranker-2BModel Scale=2B2026.06 | 74 | — | — | — | — | |
| Qwen3-VL-4B-miniRerankerModel Scale=4B, Fine-tuning Recipe=our training recipe, Reranker Type=miniReranker2026.06 | 72.5 | 68.3 | 71.5 | 73.5 | 81.8 | |
| Qwen3-VL-2B-rerankerModel Scale=2B, Fine-tuning Recipe=our training recipe2026.06 | 71.6 | 67.1 | 71.8 | 71.8 | 82.2 | |
| UME-R1Backbone=Qwen2-VL, Parameter Scale=7B2025.11 | 71.3 | 67.1 | 69.2 | 71.9 | 84.9 | |
| UME-R1Backbone=Qwen2-VL, Parameter Scale=2B, Oracle Evaluation Mode=true2025.11 | 70.2 | 67.6 | 67.5 | 71.2 | 80.1 | |
| Qwen3-VL-2B-miniRerankerModel Scale=2B, Fine-tuning Recipe=our training recipe, Reranker Type=miniReranker2026.06 | 68.2 | 65.6 | 65.7 | 69.7 | 76.5 | |
| CAFeBackbone=LLaVA-OV, Parameter Scale=7B2025.11 | 67.6 | 63.6 | 61.7 | 69.1 | 87.6 | |
| UME-R1Backbone=Qwen2-VL, Parameter Scale=2B2025.11 | 66.6 | 64.8 | 62.8 | 67.6 | 77.2 | |
| UME-R1Venue=ICLR’26, Backbone=Qwen2-VL-2B2026.04 | 66.6 | 64.8 | 62.8 | 67.6 | 77.2 | |
| DUMEBackbone=Qwen2-VL, Parameter Scale=7B2025.11 | 66.4 | 64.2 | 57 | 70.8 | 81.8 | |
| PLUMEVenue=Ours, Backbone=Qwen2-VL-2B2026.04 | 66.3 | 66.5 | 59.2 | 67.6 | 79.7 | |
| VLM2VecBackbone=Qwen2-VL, Parameter Scale=7B2025.11 | 65.5 | 62.7 | 56.9 | 69.4 | 82.2 | |
| VLM2Vec-Qwen2VL (7B)Model Scale=7B2026.06 | 65.5 | 62.7 | 56.9 | 69.4 | 82.2 | |
| VLM2Vec-V2Backbone=Qwen2-VL, Parameter Scale=2B2025.11 | 64.9 | 62.9 | 56.3 | 69.5 | 77.3 | |
| VLM2Vec-V2Venue=TMLR’26, Backbone=Qwen2-VL-2B2026.04 | 64.9 | 62 | 56.3 | 69.5 | 77.3 | |
| DUMEBackbone=Qwen2-VL, Parameter Scale=2B2025.11 | 62.5 | 59.3 | 55 | 66.3 | 78 | |
| DUMEVenue=ICLR’26, Backbone=Qwen2-VL-2B2026.04 | 62.5 | 59.3 | 55 | 66.3 | 78 | |
| Qwen3-VL-2B-InstructModel Scale=2B2026.06 | 61.7 | 59.2 | 61.2 | 60.1 | 74.2 | |
| VLM2VecBackbone=Qwen2-VL, Parameter Scale=2B2025.11 | 59.7 | 58.7 | 49.3 | 65 | 72.9 | |
| VLM2VecVenue=ICLR’25, Backbone=Qwen2-VL-2B2026.04 | 59.7 | 58.7 | 49.3 | 65 | 72.9 | |
| VLM2Vec (2B)Model Scale=2B2026.06 | 59.7 | 58.7 | 49.3 | 65 | 72.9 | |
| GMEBackbone=Qwen2-VL, Parameter Scale=7B2025.11 | 56 | 57.7 | 34.7 | 71.2 | 59.3 | |
| GME (7B)Model Scale=7B2026.06 | 56 | 57.7 | 34.7 | 71.2 | 59.3 | |
| LamRABackbone=Qwen2-VL, Parameter Scale=7B2025.11 | 54.1 | 59.2 | 26.5 | 70 | 62.7 | |
| LamRAVenue=CVPR’25, Backbone=Qwen2-VL-2B2026.04 | 54.1 | 59.2 | 26.5 | 70 | 62.7 | |
| LamRABackbone=Qwen2.5-VL, Parameter Scale=7B2025.11 | 52.4 | 51.7 | 34.1 | 66.9 | 56.7 | |
| LamRA-Qwen2.5 (7B)Model Scale=7B2026.06 | 52.4 | 51.7 | 34.1 | 66.9 | 56.7 | |
| GMEBackbone=Qwen2-VL, Parameter Scale=2B2025.11 | 51.9 | 54.4 | 29.9 | 66.9 | 55.5 | |
| GMEVenue=CVPR’25, Backbone=Qwen2-VL-2B2026.04 | 51.9 | 54.4 | 29.9 | 66.9 | 55.5 | |
| GME (2B)Model Scale=2B2026.06 | 51.9 | 54.4 | 29.9 | 66.9 | 55.5 | |
| ColPali-V1.3Backbone=PaliGemma, Parameter Scale=3B2025.11 | 34.9 | 40.3 | 11.5 | 48.1 | 40.3 | |
| ColPali v1.3 (3B)Model Scale=3B2026.06 | 34.9 | 40.3 | 11.5 | 48.1 | 40.3 |