Hallucination Evaluation on MMHal
4.2ScoreLLaVA-OneVision-7B w/ UnifiedReward-Think
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LLaVA-OneVision-7B w/ UnifiedReward-ThinkReward Model Paradigm=Generative, Base Model=LLaVA-OneVision-7B, Alignment Method=DPO2026.04 | 4.2 | — | — | — | — | — | |
| LLaVA-OneVision-7B w/ Skywork-VLReward Model Paradigm=Discriminative, Base Model=LLaVA-OneVision-7B, Alignment Method=DPO2026.04 | 4.2 | — | — | — | — | — | |
| LLaVA-OneVision-7B w/ VL-MDRReward Model Paradigm=Discriminative, Base Model=LLaVA-OneVision-7B, Alignment Method=DPO2026.04 | 4.2 | — | — | — | — | — | |
| LLaVA-OneVision-7B w/ IXC-2.5-RewardReward Model Paradigm=Discriminative, Base Model=LLaVA-OneVision-7B, Alignment Method=DPO2026.04 | 4.1 | — | — | — | — | — | |
| LLaVA-OneVision-7B w/ MM-RLHF-RewardReward Model Paradigm=Discriminative, Base Model=LLaVA-OneVision-7B, Alignment Method=DPO2026.04 | 4.1 | — | — | — | — | — | |
| CycleCapT2I Model=SD32026.03 | 4.02 | — | — | — | — | — | |
| CycleCapT2I Model=FLUX.1-dev2026.03 | 4.02 | — | — | — | — | — | |
| GPT-4o-20240513Model Name=GPT-4o-202405132024.12 | 4 | — | — | — | — | — | |
| LLaVA-OneVision-7B w/ UnifiedRewardReward Model Paradigm=Generative, Base Model=LLaVA-OneVision-7B, Alignment Method=DPO2026.04 | 4 | — | — | — | — | — | |
| CyclePrefT2I Model=SD32026.03 | 3.95 | — | — | — | — | — | |
| RICO-FlashT2I Model=FLUX.1-dev2026.03 | 3.92 | — | — | — | — | — | |
| LLaVA-OneVision-7B w/ LLaVA-CriticReward Model Paradigm=Generative, Base Model=LLaVA-OneVision-7B, Alignment Method=DPO2026.04 | 3.9 | — | — | — | — | — | |
| InternVL2.5-78BModel Name=InternVL2.5-78B2024.12 | 3.89 | — | — | — | — | — | |
| Baseline (Qwen2-VL-7B)T2I Model=—2026.03 | 3.85 | — | — | — | — | — | |
| InternVL2-Llama3-76BModel Name=InternVL2-Llama3-76B2024.12 | 3.83 | — | — | — | — | — | |
| ADAPT-TFBackbone=Qwen2.5-VL-7B2026.06 | 3.83 | — | — | — | — | 31 | |
| InternVL2-40BModel Name=InternVL2-40B2024.12 | 3.75 | — | — | — | — | — | |
| ADAPTBackbone=Qwen2.5-VL-7B2026.06 | 3.75 | — | — | — | — | 30 | |
| InternVL2.5-38BModel Name=InternVL2.5-38B2024.12 | 3.71 | — | — | — | — | — | |
| InternVL2.5-26BModel Name=InternVL2.5-26B2024.12 | 3.7 | — | — | — | — | — | |
| InternVL2.5-8BModel Name=InternVL2.5-8B2024.12 | 3.65 | — | — | — | — | — | |
| MiniCPM-V2.6Model Name=MiniCPM-V2.62024.12 | 3.6 | — | — | — | — | — | |
| DPOBackbone=Qwen2.5-VL-7B2026.06 | 3.59 | — | — | — | — | 39 | |
| InternVL2-26BModel Name=InternVL2-26B2024.12 | 3.55 | — | — | — | — | — | |
| Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B2026.06 | 3.55 | — | — | — | — | 40 | |
| VCDBackbone=Qwen2.5-VL-7B2026.06 | 3.42 | — | — | — | — | 40 | |
| Qwen2-VL-7BModel Name=Qwen2-VL-7B2024.12 | 3.4 | — | — | — | — | — | |
| InternVL2-8BModel Name=InternVL2-8B2024.12 | 3.33 | — | — | — | — | — | |
| InternVL2.5-4BModel Name=InternVL2.5-4B2024.12 | 3.31 | — | — | — | — | — | |
| LLaVA-OneVision-7BBase Model=LLaVA-OneVision-7B2026.04 | 3.2 | — | — | — | — | — | |
| ADAPTBackbone=Qwen2.5-VL-3B2026.06 | 3.16 | — | — | — | — | 37 | |
| VCDBackbone=Qwen2.5-VL-3B2026.06 | 3.14 | — | — | — | — | 42 | |
| InternVL-Chat-V1.5Model Name=InternVL-Chat-V1.52024.12 | 3.11 | — | — | — | — | — | |
| ADAPT-TFBackbone=Qwen2.5-VL-3B2026.06 | 3.11 | — | — | — | — | 39 | |
| InternVL2.5-2BModel Name=InternVL2.5-2B2024.12 | 2.94 | — | — | — | — | — | |
| TARS (Replace)Backbone=Muffin-13B2025.07 | 2.76 | — | — | 43 | — | — | |
| InternVL2-4BModel Name=InternVL2-4B2024.12 | 2.75 | — | — | — | — | — | |
| TARS (Mask)Backbone=Muffin-13B2025.07 | 2.75 | — | — | 41 | — | — | |
| CHiP-DPOBackbone=Muffin-13B2025.07 | 2.7 | — | — | 47 | — | — | |
| ADAPTBackbone=LLaVA-v1.5-13B2026.06 | 2.64 | — | — | — | — | 49 | |
| ADAPTBackbone=LLaVA-v1.5-7B2026.06 | 2.62 | — | — | — | — | 53 | |
| Qwen2.5-VL-3BBackbone=Qwen2.5-VL-3B2026.06 | 2.62 | — | — | — | — | 49 | |
| ADAPT-TFBackbone=LLaVA-v1.5-7B2026.06 | 2.55 | — | — | — | — | 53 | |
| ADAPT-TFBackbone=LLaVA-v1.5-13B2026.06 | 2.55 | — | — | — | — | 51 | |
| DPOBackbone=Qwen2.5-VL-3B2026.06 | 2.53 | — | — | — | — | 52 | |
| InternVL2-2BModel Name=InternVL2-2B2024.12 | 2.52 | — | — | — | — | — | |
| RLHF-VBackbone=LLaVA-v1.5-13B2026.06 | 2.5 | — | — | — | — | 52 | |
| InternVL2.5-1BModel Name=InternVL2.5-1B2024.12 | 2.49 | — | — | — | — | — | |
| DPOBackbone=LLaVA-v1.5-13B2026.06 | 2.47 | — | — | — | — | 51 | |
| OPERABackbone=LLaVA-v1.5-13B2026.06 | 2.46 | — | — | — | — | 52 | |
| DPOBackbone=Muffin-13B2025.07 | 2.45 | — | — | 55 | — | — | |
| mDPOBackbone=LLaVA-v1.5-7B2026.06 | 2.39 | — | — | — | — | 54 | |
| LLaVA-v1.5-13BBackbone=LLaVA-v1.5-13B2026.06 | 2.38 | — | — | — | — | 53 | |
| VCDBackbone=LLaVA-v1.5-13B2026.06 | 2.31 | — | — | — | — | 54 | |
| SIMABackbone=LLaVA-v1.5-7B2026.06 | 2.3 | — | — | — | — | — | |
| Muffin-13BBackbone=Muffin-13B2025.07 | 2.27 | — | — | 58 | — | — | |
| InternVL2-1BModel Name=InternVL2-1B2024.12 | 2.25 | — | — | — | — | — | |
| OPERABackbone=LLaVA-v1.5-7B2026.06 | 2.15 | — | — | — | — | 54 | |
| DPOBackbone=LLaVA-v1.5-7B2026.06 | 2.14 | — | — | — | — | 65 | |
| RLHFBackbone=Muffin-13B2025.07 | 2.12 | — | — | 64 | — | — | |
| VCDBackbone=LLaVA-v1.5-7B2026.06 | 2.12 | — | — | — | — | 54 | |
| LLaVA-v1.5-7BBackbone=LLaVA-v1.5-7B2026.06 | 1.86 | — | — | — | — | 64 | |
| Gemini 2.5 FlashSize=-, mode=instruct2026.04 | — | — | — | — | 4.6 | 23.9 | |
| InternVL3.5Size=8B, mode=instruct2026.04 | — | — | — | — | 3.8 | 34.7 | |
| LLaVA2025.04 | — | 31.9 | 70.8 | — | — | — | |
| MiniCPM-o 4.5Size=9B, mode=instruct2026.04 | — | — | — | — | 4.7 | 24.3 | |
| POPEN2025.04 | — | 42.1 | 47.9 | — | — | — | |
| Qwen3-OmniSize=30B-A3B, mode=instruct2026.04 | — | — | — | — | 4.6 | 31.6 | |
| Qwen3-VLSize=8B, mode=instruct2026.04 | — | — | — | — | 4.7 | 29.9 | |
| RLHF-V2025.04 | — | 40 | 52.1 | — | — | — | |
| SFT + DA-DPOBase Model=LLaVA-OneVision-7B [3]2026.06 | — | — | — | — | — | 2.78 | |
| SFT + DA-DPOBase Model=InternVL2.5-26B [4]2026.06 | — | — | — | — | — | 2.4 | |
| SFT + DPOBase Model=LLaVA-OneVision-7B [3]2026.06 | — | — | — | — | — | 2.61 | |
| SFT + DPOBase Model=InternVL2.5-26B [4]2026.06 | — | — | — | — | — | 2.35 | |
| SFT + VIGIL (Ours)Base Model=LLaVA-OneVision-7B [3]2026.06 | — | — | — | — | — | 2.45 | |
| SFT + VIGIL (Ours)Base Model=InternVL2.5-26B [4]2026.06 | — | — | — | — | — | 2.15 | |
| SFT onlyBase Model=LLaVA-OneVision-7B [3]2026.06 | — | — | — | — | — | 2.76 | |
| SFT onlyBase Model=InternVL2.5-26B [4]2026.06 | — | — | — | — | — | 2.5 |