Multimodal Embedding on MMEB
76.1Classification AccuracySeed-1.6-embedding
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Seed-1.6-embeddingBackbone=Seed1.6-flash, Model Size=unknown2026.03 | 76.1 | 74 | 77.9 | 91.3 | — | — | 77.8 | — | — | — | |
| Qwen3-VL-EmbeddingBackbone=Qwen3-VL, Model Size=8.14B2026.03 | 74.2 | 81.1 | 80.2 | 92.3 | 81.9 | 78 | 80.1 | — | — | — | |
| MetaEmbed-32BSize=32B, Initialization=Qwen2.5-VL2025.09 | 73.7 | 78.6 | 78.9 | 88.1 | 82.8 | 73.7 | 78.7 | — | — | — | |
| MetaEmbed-7BSize=7B, Initialization=Qwen2.5-VL2025.09 | 71.3 | 74.2 | 78.7 | 85.4 | 81.8 | 70 | 76.6 | — | — | — | |
| TSEmbedBackbone=Qwen2-VL, Model Size=8.40B2026.03 | 71.1 | 70.3 | 75.9 | 91.3 | 78.8 | 69.6 | 74.7 | — | — | — | |
| RzenEmbedBackbone=Qwen2-VL, Model Size=8.29B2026.03 | 70.6 | 71.7 | 78.5 | 92.1 | 78.5 | 72.7 | 75.9 | — | — | — | |
| Qwen3-VL-EmbeddingBackbone=Qwen3-VL, Model Size=2.13B2026.03 | 70.3 | 74.3 | 74.8 | 88.5 | 75.3 | 74.5 | 75 | — | — | — | |
| B3Backbone=Qwen2-VL, Model Size=8.29B2026.03 | 70 | 66.5 | 74.1 | 84.6 | 75.9 | 67.1 | 72 | — | — | — | |
| B3-7BSize=7B2025.09 | 70 | 66.5 | 74.1 | 84.6 | 75.9 | 67.1 | 72 | — | — | — | |
| Ops-MM-embedding-v1Backbone=Qwen2-VL, Model Size=8.29B2026.03 | 69.7 | 69.6 | 73.1 | 87.2 | — | — | 72.7 | — | — | — | |
| TSEmbedBackbone=Qwen2-VL, Model Size=2.26B2026.03 | 68.8 | 64.3 | 72.1 | 85.7 | 74.5 | 65.6 | 70.5 | — | — | — | |
| RzenEmbedBackbone=Qwen2-VL, Model Size=2.21B2026.03 | 68.5 | 66.3 | 74.5 | 90.3 | 76.1 | 67.4 | 72.3 | — | — | — | |
| UNITEBackbone=Qwen2-VL, Model Size=8.29B2026.03 | 68.3 | 65.1 | 71.6 | 84.8 | 73.6 | 66.3 | 70.3 | — | — | — | |
| Ops-MM-embedding-v1Backbone=Qwen2-VL, Model Size=2.21B2026.03 | 68.1 | 65.1 | 69.2 | 80.9 | — | — | 69 | — | — | — | |
| mmE5Size=11B2025.09 | 67.6 | 62.7 | 71 | 89.7 | 72.4 | 66.6 | 69.8 | — | — | — | |
| B3Backbone=Qwen2-VL, Model Size=2.21B2026.03 | 67 | 61.2 | 70.9 | 79.9 | 72.1 | 63.1 | 68.1 | — | — | — | |
| QQMM-embedBackbone=LLaVA-OV, Model Size=8.297B2026.03 | 66.8 | 66.8 | 70.5 | 90.4 | 74.7 | 65.6 | 70.7 | — | — | — | |
| MetaEmbed-11BSize=11B, Initialization=Llama-3.2-Vision2025.09 | 66.4 | 42.1 | 74.3 | 91.6 | 65.7 | 64.3 | 65.1 | — | — | — | |
| MoCa-7BSize=7B2025.09 | 65.8 | 64.7 | 75 | 92.4 | 74.7 | 67.6 | 71.5 | — | — | — | |
| LLaVEBackbone=LLaVA-OV, Model Size=8.03B2026.03 | 65.7 | 65.4 | 70.9 | 91.9 | 75 | 64.4 | 70.3 | — | — | — | |
| UniME-V2Backbone=LLaVA-OV, Model Size=8.03B2026.03 | 65.3 | 67.6 | 72.9 | 90.2 | 74.8 | 66.7 | 71.2 | — | — | — | |
| CAFeBackbone=LLaVA-OV, Model Size=8.03B2026.03 | 65.2 | 65.6 | 70 | 91.2 | 75.8 | 62.4 | 69.8 | — | — | — | |
| MetaEmbed-3BGemmaSize=3B, Initialization=PaliGemma2025.09 | 64.9 | 53.5 | 70.9 | 79.5 | 68.6 | 61.3 | 65.4 | — | — | — | |
| UniME-V2Backbone=Qwen2-VL, Model Size=8.29B2026.03 | 64 | 60.1 | 73.1 | 82.8 | 72 | 63 | 68 | — | — | — | |
| UNITEBackbone=Qwen2-VL, Model Size=2.21B2026.03 | 63.2 | 55.9 | 65.4 | 75.6 | 65.8 | 60.1 | 63.3 | — | — | — | |
| LoRA-αBackbone=Qwen 2-VL, Model Scale=7B, Rank (r)=16, Learning Rate (η)=2 × 10−52026.06 | 63 | 59.4 | 70.6 | 84.2 | 73.3 | 58.8 | 66.9 | — | — | — | |
| VLM2VEC-V2Backbone=Qwen2-VL, Model Size=2.21B2026.03 | 62.9 | 56.3 | 69.5 | 77.3 | — | — | 64.9 | — | — | — | |
| VLM2Vec-V2Size=2B2025.09 | 62.9 | 56.3 | 69.5 | 77.3 | — | — | 64.9 | — | — | — | |
| MetaEmbed-3BSize=3B, Initialization=Qwen2.5-VL2025.09 | 62.7 | 68.1 | 71.9 | 79.6 | 73.5 | 63.8 | 69.1 | — | — | — | |
| VLM2VECBackbone=Qwen2-VL, Model Size=8.29B2026.03 | 62.6 | 57.8 | 69.9 | 81.7 | 65.2 | 56.3 | 65.8 | — | — | — | |
| UniMoCoFine-tuning protocol=MMEB, Model Backbone=Qwen2-VL-7B2025.05 | 62.6 | 55.5 | 65 | 78.2 | 67 | 58.4 | 63.2 | 59.6 | 73.6 | 65.1 | |
| VLM2VecBackbone=Qwen 2-VL, Model Scale=7B2026.06 | 62.6 | 57.8 | 69.9 | 81.7 | 72.7 | 57.8 | 65.8 | — | — | — | |
| UniME-V2Backbone=Qwen2-VL, Model Size=2.21B2026.03 | 62.1 | 56.3 | 68 | 72.7 | 67.4 | 58.9 | 63.6 | — | — | — | |
| LLaVEBackbone=Aquila-VL, Model Size=1.95B2026.03 | 62.1 | 60.2 | 65.2 | 84.9 | 69.4 | 59.8 | 65.2 | — | — | — | |
| LoRA (10× LR)Backbone=Qwen 2-VL, Model Scale=7B, Rank (r)=16, Learning Rate (η)=2 × 10−42026.06 | 62 | 58.9 | 69.3 | 81.3 | 72.6 | 55.4 | 65.3 | — | — | — | |
| DoRABackbone=Qwen 2-VL, Model Scale=7B, Rank (r)=16, Learning Rate (η)=2 × 10−52026.06 | 61.6 | 54.2 | 66.5 | 78.2 | 69 | 55.6 | 63 | — | — | — | |
| VLM2VECBackbone=LLaVA-1.6, Model Size=7.57B2026.03 | 61.2 | 49.9 | 67.4 | 86.1 | 67.5 | 57.1 | 62.9 | — | — | — | |
| VLM2VecSize=7B2025.09 | 61.2 | 49.9 | 67.4 | 86.1 | 67.5 | 57.1 | 62.9 | — | — | — | |
| LoRABackbone=Qwen 2-VL, Model Scale=7B, Rank (r)=16, Learning Rate (η)=2 × 10−52026.06 | 60.7 | 57.3 | 68.1 | 81.2 | 70.7 | 56.8 | 64.5 | — | — | — | |
| LoRA-αBackbone=Qwen 2-VL, Model Scale=2B, Rank (r)=16, Learning Rate (η)=2 × 10−52026.06 | 60.2 | 52.3 | 66 | 75.2 | 67.2 | 54.7 | 61.6 | — | — | — | |
| ABCSize=7B2025.09 | 60 | 31 | — | — | — | — | — | — | — | — | |
| MoCa-3BSize=3B2025.09 | 59.8 | 62.9 | 70.6 | 88.6 | 72.3 | 61.5 | 67.5 | — | — | — | |
| LamRABackbone=Qwen2-VL, Model Size=8.29B2026.03 | 59.2 | 26.5 | 70 | 62.7 | — | — | 54.1 | — | — | — | |
| CAFeBackbone=LLaVA-OV, Model Size=0.894B2026.03 | 59.1 | 49.1 | 61 | 83 | 64.3 | 53.7 | 59.6 | — | — | — | |
| VLM2VECBackbone=Qwen2-VL, Model Size=2.21B2026.03 | 59 | 49.4 | 65.4 | 73.4 | 66 | 52.6 | 59.3 | — | — | — | |
| VLM2VecBackbone=Qwen 2-VL, Model Scale=2B2026.06 | 59 | 49.4 | 65.4 | 73.4 | 66 | 52.6 | 60.1 | — | — | — | |
| LoRA (10× LR)Backbone=Qwen 2-VL, Model Scale=2B, Rank (r)=16, Learning Rate (η)=2 × 10−42026.06 | 59 | 50.9 | 65.8 | 73 | 66.9 | 52.5 | 60.5 | — | — | — | |
| LoRABackbone=Qwen 2-VL, Model Scale=2B, Rank (r)=16, Learning Rate (η)=2 × 10−52026.06 | 58.7 | 47.2 | 64.5 | 70.5 | 64 | 52.1 | 58.7 | — | — | — | |
| Full Fine-TuningBackbone=Qwen 2-VL, Model Scale=2B, Learning Rate (η)=2 × 10−52026.06 | 58.3 | 48.9 | 64.8 | 74.5 | 67.2 | 50.3 | 59.6 | — | — | — | |
| PiSSABackbone=Qwen 2-VL, Model Scale=7B, Rank (r)=16, Learning Rate (η)=2 × 10−52026.06 | 58.2 | 55 | 65.4 | 77.9 | 67.9 | 54.4 | 61.9 | — | — | — | |
| GMEBackbone=Qwen2-VL, Model Size=8.29B2026.03 | 57.7 | 34.7 | 71.2 | 59.3 | — | — | 56 | — | — | — | |
| DoRABackbone=Qwen 2-VL, Model Scale=2B, Rank (r)=16, Learning Rate (η)=2 × 10−52026.06 | 57.3 | 43.2 | 60.8 | 67 | 61.3 | 48.6 | 55.6 | — | — | — | |
| GMESize=7B2025.09 | 56.9 | 41.2 | 67.8 | 53.4 | — | — | 55.8 | — | — | — | |
| MMRetSize=7B2025.09 | 56 | 57.4 | 69.9 | 83.6 | 68 | 59.1 | 64.1 | — | — | — | |
| CLIPSize=428M2025.09 | 55.2 | 19.7 | 53.2 | 62.2 | 47.6 | 42.8 | 45.4 | — | — | — | |
| UniMoCoFine-tuning protocol=MMEB, Model Backbone=Phi-3.5V2025.05 | 55 | 58.2 | 63.2 | 82.4 | 68.2 | 53.5 | 61.7 | 57.7 | 72.8 | 64.1 | |
| VLM2VECBackbone=Phi-3.5-V, Model Size=4.15B2026.03 | 54.8 | 54.9 | 62.3 | 79.5 | 66.5 | 52 | 60.1 | — | — | — | |
| VLM2VECFine-tuning protocol=MMEB, Model Backbone=Phi-3.5V2025.05 | 54.8 | 54.9 | 62.3 | 79.5 | 66.5 | 52 | 60.1 | 56.1 | 72.6 | 61.5 | |
| GMEBackbone=Qwen2-VL, Model Size=2.21B2026.03 | 54.4 | 29.9 | 66.9 | 55.5 | — | — | 51.9 | — | — | — | |
| PiSSABackbone=Qwen 2-VL, Model Scale=2B, Rank (r)=16, Learning Rate (η)=2 × 10−52026.06 | 52.5 | 44.2 | 61.4 | 66.1 | 59.3 | 48.9 | 54.7 | — | — | — | |
| LamRABackbone=Qwen2.5-VL, Model Size=8.29B2026.03 | 51.7 | 34.1 | 66.9 | 56.7 | — | — | 52.4 | — | — | — | |
| OpenCLIPFine-tuning protocol=MMEB2025.05 | 51.4 | 30.8 | 58.1 | 66.3 | 56.9 | 40.4 | 49.6 | 43.1 | 66.8 | 54.4 | |
| CLIPFine-tuning protocol=MMEB2025.05 | 50 | 27 | 55.3 | 64.8 | 52.2 | 38.9 | 47 | 40.3 | 64.7 | 49.3 | |
| MM-EMBEDSize=7B2025.09 | 48.1 | 32.2 | 63.8 | 57.8 | — | — | 50 | — | — | — | |
| OpenCLIPBackbone=-, Model Size=0.428B2026.03 | 47.8 | 10.9 | 52.3 | 53.3 | 39.3 | 40.2 | 39.7 | — | — | — | |
| OpenCLIPFine-tuning protocol=None2025.05 | 47.8 | 10.9 | 52.3 | 53.3 | 39.3 | 40.2 | 39.7 | 33.1 | 57.9 | 44.2 | |
| UniIR (CLIP_SF)Backbone=-, Model Size=0.428B2026.03 | 44.3 | 16.2 | 61.8 | 65.3 | 47.1 | 41.7 | 44.7 | — | — | — | |
| CLIPBackbone=-, Model Size=0.428B2026.03 | 42.8 | 9.1 | 53 | 51.8 | 37.1 | 38.7 | 37.8 | — | — | — | |
| CLIPFine-tuning protocol=None2025.05 | 42.8 | 9.1 | 53 | 51.8 | 37.1 | 38.7 | 37.8 | 29.8 | 62.1 | 41.6 | |
| UniIR (BLIP_FF)Backbone=-, Model Size=0.247B2026.03 | 42.1 | 15 | 60.1 | 62.2 | 44.7 | 40.4 | 42.8 | — | — | — | |
| UniIRSize=428M2025.09 | 42.1 | 15 | 60.1 | 62.2 | — | — | 42.8 | — | — | — | |
| UniIRFine-tuning protocol=None2025.05 | 42.1 | 15 | 60.1 | 62.2 | 44.7 | 40.4 | 42.8 | 32.5 | 58.2 | 59.7 | |
| SigLIPBackbone=-, Model Size=0.203B2026.03 | 40.3 | 8.4 | 31.6 | 59.5 | 32.3 | 38 | 34.8 | — | — | — | |
| ColPali-v1.3Backbone=PaliGemma, Model Size=2.92B2026.03 | 40.3 | 11.5 | 48.1 | 40.3 | — | — | 34.9 | — | — | — | |
| SigLIPFine-tuning protocol=None2025.05 | 40.3 | 8.4 | 31.6 | 59.5 | 32.3 | 38 | 34.8 | 27 | 44.6 | 49 | |
| MagiclensBackbone=-, Model Size=0.428B2026.03 | 38.8 | 8.3 | 35.4 | 26 | 31 | 23.7 | 27.8 | — | — | — | |
| MagicLensSize=613M2025.09 | 38.8 | 8.3 | 35.4 | 26 | — | — | 27.8 | — | — | — | |
| BLIP-2Backbone=-, Model Size=3.74B2026.03 | 27 | 4.2 | 33.9 | 47 | 25.3 | 25.1 | 25.2 | — | — | — | |
| BLIP2Fine-tuning protocol=None2025.05 | 27 | 4.2 | 33.9 | 47 | 25.3 | 25.1 | 25.2 | 15.7 | 43.1 | 37.9 |