Science Question Answering on ScienceQA (Component Metrics)
88.6IMG ScoreQwen2.5-VL-7B-Instruct
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Qwen2.5-VL-7B-InstructMethod=baseline2026.02 | 88.6 | — | — | — | — | — | — | — | — | |
| HII-DPOModel=Qwen2.5-VL-7B-Instruct2026.02 | 88.5 | — | — | — | — | — | — | — | — | |
| Human#T-Param=-2026.03 | 87.5 | 90.23 | 84.97 | 87.48 | 89.6 | 88.1 | 91.59 | 82.42 | 88.4 | |
| Qwen2.5-VL-7B +Hit-RAG#T-Param=7B2026.03 | 87.41 | 92.58 | 94.83 | 92.27 | 81.23 | 97.28 | 87.67 | 86.82 | 92.97 | |
| HII-DPOModel=Qwen2-VL-7B-Instruct2026.02 | 86.7 | — | — | — | — | — | — | — | — | |
| Qwen2-VL-7B-InstructMethod=baseline2026.02 | 85.7 | — | — | — | — | — | — | — | — | |
| Qwen2.5-72B+Hit-RAG#T-Param=72B2026.03 | 80.76 | 87.61 | 89.43 | 95.91 | 86.07 | 97.63 | 90.86 | 84.11 | 90.14 | |
| Qwen2.5-7B+Hit-RAG#T-Param=4B2026.03 | 78.78 | 86.59 | 82 | 90.36 | 83.63 | 92.4 | 88.66 | 81.02 | 86.61 | |
| Qwen2.5-VL-7B#T-Param=7B2026.03 | 78.73 | 83.7 | 80.31 | 91.91 | 82.84 | 89.2 | 86.71 | 84.05 | 85.12 | |
| Qwen2-VL-2B-InstructMethod=baseline2026.02 | 76.9 | — | — | — | — | — | — | — | — | |
| HII-DPOModel=Qwen2-VL-2B-Instruct2026.02 | 76.9 | — | — | — | — | — | — | — | — | |
| Qwen2.5-72B + VaLiK#T-Param=72B2026.03 | 74.17 | 85.61 | 75.93 | 90.27 | 84.4 | 92.33 | 85.79 | 82.98 | 84.77 | |
| Qwen2.5-7B + VaLiK#T-Param=7B2026.03 | 73.18 | 84.15 | 75.14 | 87.64 | 82.99 | 89.69 | 84.4 | 80.95 | 83.16 | |
| SENTINELModel=LLaVA-v1.5-13B2026.02 | 72.8 | — | — | — | — | — | — | — | — | |
| HII-DPOModel=LLaVA-v1.5-13B2026.02 | 72.8 | — | — | — | — | — | — | — | — | |
| DDCoT#T-Param=175B2026.03 | 72.53 | 80.15 | 76.72 | 82.82 | 78.89 | 85.02 | 82.86 | 75.21 | 80.15 | |
| vanilla-DPOModel=LLaVA-v1.5-13B2026.02 | 71.8 | — | — | — | — | — | — | — | — | |
| baselineModel=LLaVA-v1.5-13B2026.02 | 71.6 | — | — | — | — | — | — | — | — | |
| CoT (GPT-4)#T-Param=1T+2026.03 | 71.49 | 85.48 | 72.44 | 90.27 | 82.65 | 92.89 | 86.66 | 79.04 | 83.99 | |
| HSA-DPOModel=LLaVA-v1.5-13B2026.02 | 71.3 | — | — | — | — | — | — | — | — | |
| Chameleon (ChatGPT)#T-Param=175B+2026.03 | 70.8 | 81.62 | 70.64 | 84 | 79.77 | 86.62 | 81.86 | 76.53 | 79.93 | |
| GPT-4#T-Param=-2026.03 | 70.75 | 84.06 | 73.45 | 87.36 | 81.87 | 90.73 | 84.69 | 79.1 | 82.69 | |
| HA-DPOModel=LLaVA-v1.5-7B2026.02 | 69.7 | — | — | — | — | — | — | — | — | |
| mDPOModel=LLaVA-v1.5-13B2026.02 | 69.4 | — | — | — | — | — | — | — | — | |
| SENTINELModel=LLaVA-v1.5-7B2026.02 | 69.2 | — | — | — | — | — | — | — | — | |
| Re-AlignModel=LLaVA-v1.5-13B2026.02 | 69 | — | — | — | — | — | — | — | — | |
| POVIDModel=LLaVA-v1.5-7B2026.02 | 68.8 | — | — | — | — | — | — | — | — | |
| HII-DPOModel=LLaVA-v1.5-7B2026.02 | 68.6 | — | — | — | — | — | — | — | — | |
| RLAIF-VModel=LLaVA-v1.5-7B2026.02 | 68.2 | — | — | — | — | — | — | — | — | |
| CoT (GPT-3)#T-Param=173B2026.03 | 67.43 | 75.44 | 70.87 | 78.09 | 74.68 | 79.93 | 78.23 | 69.68 | 75.17 | |
| mDPOModel=LLaVA-v1.5-7B2026.02 | 67.3 | — | — | — | — | — | — | — | — | |
| TPOModel=LLaVA-v1.5-7B2026.02 | 67.1 | — | — | — | — | — | — | — | — | |
| baselineModel=LLaVA-v1.5-7B2026.02 | 66.8 | — | — | — | — | — | — | — | — | |
| Re-AlignModel=LLaVA-v1.5-7B2026.02 | 66.6 | — | — | — | — | — | — | — | — | |
| CoT (UnifiedQA)#T-Param=223M2026.03 | 66.53 | 71 | 76.04 | 78.91 | 66.42 | 81.81 | 77.06 | 68.82 | 74.11 | |
| Qwen2.5-7B#T-Param=7B2026.03 | 65.79 | 76.2 | 67.83 | 77.27 | 74.49 | 79.02 | 77.72 | 69.35 | 74.72 | |
| Qwen2.5-72B#T-Param=72B2026.03 | 65 | 79.64 | 67.1 | 84.9 | 77.56 | 87.93 | 80.25 | 74.85 | 78.37 | |
| MIND_baseLearning=Fine-tune, Size=223M2025.12 | 0.9276 | 0.9307 | 0.9674 | 0.8709 | 0.9242 | 0.8927 | 0.9291 | 0.9117 | 0.9229 | |
| DPMM-CoT_baseLearning=Fine-tune, Size=306M2025.12 | 0.9048 | 0.9272 | 0.8785 | 0.8991 | 0.9272 | 0.9129 | 0.9145 | 0.9011 | 0.9097 | |
| Multimodal-T-SciQ_baseLearning=Fine-tune, Size=223M2025.12 | 0.9033 | 0.9152 | 0.9145 | 0.9245 | 0.9194 | 0.9226 | 0.9211 | 0.911 | 0.9175 | |
| Gemini+RMRLearning=Few-shot2025.12 | 0.8969 | 0.9179 | 0.9426 | 0.8964 | 0.914 | 0.9101 | 0.9284 | 0.8978 | 0.9175 | |
| MC-CoT_baseLearning=Fine-tune, Size=223M2025.12 | 0.883 | 0.9187 | 0.8459 | 0.93 | 0.9228 | 0.9275 | 0.9064 | 0.9064 | 0.9064 | |
| LLaVALearning=Fine-tune, Size=13B2025.12 | 0.88 | 0.9036 | 0.9595 | 0.88 | 0.8949 | 0.9066 | 0.9093 | 0.909 | 0.9092 | |
| Human2025.12 | 0.875 | 0.9023 | 0.8497 | 0.8748 | 0.896 | 0.881 | 0.9159 | 0.8242 | 0.884 | |
| LaVINLearning=Fine-tune, Size=7B2025.12 | 0.8746 | 0.8925 | 0.9494 | 0.8524 | 0.8851 | 0.8808 | 0.9016 | 0.8807 | 0.8941 | |
| DDCoT_baseLearning=Fine-tune, Size=223M2025.12 | 0.8334 | 0.8872 | 0.8684 | 0.8491 | 0.8759 | 0.8808 | 0.8858 | 0.851 | 0.8734 | |
| Multimodal-CoT_baseLearning=Fine-tune, Size=223M2025.12 | 0.8275 | 0.8406 | 0.9235 | 0.8218 | 0.8275 | 0.8474 | 0.8579 | 0.8444 | 0.8531 | |
| LLaMA-AdapterLearning=Fine-tune, Size=6B2025.12 | 0.8032 | 0.8437 | 0.883 | 0.8436 | 0.8372 | 0.869 | 0.8583 | 0.8405 | 0.8519 | |
| Chameleon (GPT-4)Learning=Few-shot2025.12 | 0.7764 | 0.8983 | 0.7413 | 0.8982 | 0.8827 | 0.9213 | 0.8803 | 0.8372 | 0.8654 | |
| GPT-4 w/ coTLearning=Few-shot2025.12 | 0.7149 | 0.8548 | 0.7244 | 0.9027 | 0.8265 | 0.9289 | 0.8666 | 0.7904 | 0.8399 | |
| Chameleon (ChatGPT)Learning=Few-shot2025.12 | 0.708 | 0.8162 | 0.7064 | 0.84 | 0.7977 | 0.8662 | 0.8186 | 0.7653 | 0.7993 | |
| ChatGPT w/ coTLearning=Few-shot, Size=175B2025.12 | 0.6792 | 0.7882 | 0.7098 | 0.8318 | 0.7737 | 0.8613 | 0.8072 | 0.7403 | 0.7831 | |
| GPT-3.5 w/ coTLearning=Few-shot, Size=175B2025.12 | 0.6743 | 0.7544 | 0.7087 | 0.7809 | 0.7468 | 0.7993 | 0.7823 | 0.6968 | 0.7517 | |
| GPT-3.5Learning=Few-shot, Size=175B2025.12 | 0.6728 | 0.7464 | 0.6974 | 0.76 | 0.7444 | 0.7742 | 0.768 | 0.6889 | 0.7397 | |
| UnifiedQA_base w/ coTLearning=Fine-tune, Size=223M2025.12 | 0.6653 | 0.71 | 0.7604 | 0.7891 | 0.6642 | 0.8181 | 0.7706 | 0.6882 | 0.7411 | |
| VisualBERTLearning=Fine-tune, Size=111M2025.12 | 0.6217 | 0.5933 | 0.6918 | 0.6118 | 0.6271 | 0.5854 | 0.6296 | 0.5992 | 0.6187 | |
| ViLTLearning=Fine-tune, Size=113M2025.12 | 0.6138 | 0.6048 | 0.6389 | 0.6027 | 0.632 | 0.57 | 0.6072 | 0.619 | 0.6114 | |
| UnifiedQA_baseLearning=Fine-tune, Size=223M2025.12 | 0.6138 | 0.6816 | 0.6918 | 0.7491 | 0.6378 | 0.7784 | 0.7298 | 0.65 | 0.7012 | |
| Patch-TRMLearning=Fine-tune, Size=90M2025.12 | 0.5528 | 0.6519 | 0.4679 | 0.6555 | 0.6696 | 0.6495 | 0.5804 | 0.675 | 0.6142 | |
| Top-DownLearning=Fine-tune, Size=70M2025.12 | 0.5488 | 0.595 | 0.5433 | 0.6182 | 0.629 | 0.5979 | 0.5727 | 0.6216 | 0.5902 | |
| DFAFLearning=Fine-tune, Size=74M2025.12 | 0.5449 | 0.6403 | 0.4882 | 0.6355 | 0.6588 | 0.6411 | 0.5712 | 0.6717 | 0.6072 | |
| BANLearning=Fine-tune, Size=112M2025.12 | 0.526 | 0.6088 | 0.4657 | 0.6664 | 0.6261 | 0.6551 | 0.5683 | 0.6394 | 0.5937 | |
| MCANLearning=Fine-tune, Size=95M2025.12 | 0.5117 | 0.5608 | 0.4623 | 0.5809 | 0.5943 | 0.554 | 0.5165 | 0.5972 | 0.5454 | |
| Random2025.12 | 0.4008 | 0.4028 | 0.4613 | 0.2925 | 0.4745 | 0.3366 | 0.3935 | 0.4067 | 0.3983 | |
| BLIP-2#T-Param=-2026.03 | — | — | — | — | — | — | — | — | 74.17 | |
| CCOT#T-Param=7B2026.03 | — | — | — | — | — | — | — | — | 76.84 | |
| GraphVis#T-Param=7B2026.03 | — | — | — | — | — | — | — | — | 73.18 | |
| LaVIN-13B#T-Param=-2026.03 | — | — | — | — | — | — | — | — | 77.54 | |
| LG-VQA (BLIP-2)#T-Param=-2026.03 | — | — | — | — | — | — | — | — | 86.32 |