Hallucination Assessment on HallusionBench
71.6Answer Accuracy (aAcc)ResDec
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ResDecBackbone=Qwen2.5-VL, Max New Tokens=1282026.02 | 71.6 | 49 | 47.1 | — | — | |
| DefenderIteration=32026.01 | 69.19 | — | — | — | — | |
| RegularBackbone=Qwen2.5-VL, Max New Tokens=1282026.02 | 69.1 | 45.8 | 43.4 | — | — | |
| OPERABackbone=Qwen2.5-VL, Max New Tokens=1282026.02 | 68.7 | 43.4 | 41.5 | — | — | |
| VCDBackbone=Qwen2.5-VL, Max New Tokens=1282026.02 | 68.7 | 41.3 | 39.8 | — | — | |
| Liu et al.Augmentation=Insert2026.01 | 68.45 | — | — | — | — | |
| Yang et al.2026.01 | 68.35 | — | — | — | — | |
| DefenderIteration=22026.01 | 68.24 | — | — | — | — | |
| Ablation (Clean Data)2026.01 | 67.93 | — | — | — | — | |
| DefenderIteration=12026.01 | 67.82 | — | — | — | — | |
| Liu et al.Augmentation=Add2026.01 | 67.61 | — | — | — | — | |
| Liu et al.Augmentation=All2026.01 | 67.61 | — | — | — | — | |
| ICDBackbone=Qwen2.5-VL, Max New Tokens=1282026.02 | 67.6 | 39.8 | 38.7 | — | — | |
| Base (M def (0))2026.01 | 67.51 | — | — | — | — | |
| DoLaBackbone=Qwen2.5-VL, Max New Tokens=1282026.02 | 66.9 | 39.5 | 40.8 | — | — | |
| MM-EurekaClueTracer=true2026.02 | 65.1 | 33.41 | 47.11 | 56.98 | — | |
| Ocean-R1ClueTracer=true2026.02 | 63.51 | 30.11 | 43.35 | 54.88 | — | |
| ORSTA-R1ClueTracer=true2026.02 | 60.05 | 27.91 | 36.71 | 53.49 | — | |
| R1-OneVisionClueTracer=true2026.02 | 58.9 | 24.18 | 40.17 | 54.42 | — | |
| MM-EurekaClueTracer=false2026.02 | 58.46 | 23.07 | 31.5 | 49.3 | — | |
| ORSTA-R1ClueTracer=false2026.02 | 55.45 | 21.76 | 28.61 | 45.81 | — | |
| Ocean-R1ClueTracer=false2026.02 | 48.89 | 19.34 | 27.76 | 40 | — | |
| ResDecBackbone=InstructBLIP, Max New Tokens=1282026.02 | 46.5 | 10.2 | 11.3 | — | — | |
| AGLABackbone=InstructBLIP, Max New Tokens=1282026.02 | 45.9 | 9.2 | 9.7 | — | — | |
| RegularBackbone=InstructBLIP, Max New Tokens=1282026.02 | 45.3 | 9.5 | 10.1 | — | — | |
| ResDecBackbone=LLaVA-1.5, Max New Tokens=1282026.02 | 45.1 | 12.7 | 18.2 | — | — | |
| DoLaBackbone=InstructBLIP, Max New Tokens=1282026.02 | 43.8 | 8.6 | 8.4 | — | — | |
| VCDBackbone=InstructBLIP, Max New Tokens=1282026.02 | 43.3 | 8.5 | 8.1 | — | — | |
| AGLABackbone=LLaVA-1.5, Max New Tokens=1282026.02 | 42.9 | 8.3 | 17.2 | — | — | |
| ONLYBackbone=LLaVA-1.5, Max New Tokens=1282026.02 | 42.7 | 9.3 | 17.5 | — | — | |
| MemVRBackbone=LLaVA-1.5, Max New Tokens=1282026.02 | 42.5 | 9 | 17.9 | — | — | |
| OPERABackbone=InstructBLIP, Max New Tokens=1282026.02 | 42.3 | 8.2 | 9.7 | — | — | |
| RegularBackbone=LLaVA-1.5, Max New Tokens=1282026.02 | 41.5 | 8.1 | 17.9 | — | — | |
| DoLaBackbone=LLaVA-1.5, Max New Tokens=1282026.02 | 41.5 | 9.3 | 15.7 | — | — | |
| OPERABackbone=LLaVA-1.5, Max New Tokens=1282026.02 | 41.2 | 5.5 | 16.2 | — | — | |
| VCDBackbone=LLaVA-1.5, Max New Tokens=1282026.02 | 41.1 | 11.4 | 13.9 | — | — | |
| ICDBackbone=InstructBLIP, Max New Tokens=1282026.02 | 41 | 7.5 | 7.2 | — | — | |
| ICDBackbone=LLaVA-1.5, Max New Tokens=1282026.02 | 38.2 | 8.4 | 13.9 | — | — | |
| R1-OneVisionClueTracer=false2026.02 | 35.43 | 8.57 | 12.43 | 31.86 | — | |
| ACPOModel Scale=14B, Base Model=InternVL32026.03 | — | — | — | — | 70 | |
| ACPOModel Scale=8B, Base Model=InternVL32026.03 | — | — | — | — | 69.1 | |
| BAGELModel Category=Unified Models2026.04 | — | — | — | — | 52.5 | |
| DPOModel Scale=14B, Base Model=InternVL32026.03 | — | — | — | — | 69.7 | |
| DPOModel Scale=8B, Base Model=InternVL32026.03 | — | — | — | — | 67 | |
| DPO-ShiftModel Scale=14B, Base Model=InternVL32026.03 | — | — | — | — | 70 | |
| DPO-ShiftModel Scale=8B, Base Model=InternVL32026.03 | — | — | — | — | 65.2 | |
| InstructModel Scale=14B, Base Model=InternVL32026.03 | — | — | — | — | 68.9 | |
| InstructModel Scale=8B, Base Model=InternVL32026.03 | — | — | — | — | 66.8 | |
| InternVL-UModel Category=Unified Models2026.04 | — | — | — | — | 44.8 | |
| IPOModel Scale=14B, Base Model=InternVL32026.03 | — | — | — | — | 69.8 | |
| IPOModel Scale=8B, Base Model=InternVL32026.03 | — | — | — | — | 62.8 | |
| LLaDA-oModel Category=Unified Models2026.04 | — | — | — | — | 47.4 | |
| LLaDA-VModel Category=Specialist VLMs2026.04 | — | — | — | — | 39.2 | |
| LLaDA2.0-UniModel Category=Unified Models2026.04 | — | — | — | — | 50.2 | |
| Lumina-DiMOOModel Category=Unified Models2026.04 | — | — | — | — | 32.9 | |
| Qwen2.5-VL-7BModel Category=Specialist VLMs2026.04 | — | — | — | — | 51.9 | |
| SimPOModel Scale=14B, Base Model=InternVL32026.03 | — | — | — | — | 68 | |
| SimPOModel Scale=8B, Base Model=InternVL32026.03 | — | — | — | — | 66.1 | |
| β-DPOModel Scale=14B, Base Model=InternVL32026.03 | — | — | — | — | 69.5 | |
| β-DPOModel Scale=8B, Base Model=InternVL32026.03 | — | — | — | — | 65.7 |