Visual Question Answering on Chest X-ray VQA (test)
96.49Overall AccuracyOracle
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| OracleType=Selection Method2026.02 | 96.49 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CheXPO-v2-Phi4MM(5k)Model Category=Our Proposal, Optimization=Knowledge Graph Consistency GRPO, Sample Size=5k, Backbone=Phi-4MM2025.12 | 86 | 92.07 | 85.93 | 76.36 | 52 | 100 | 72.73 | 100 | 75.41 | 92.31 | 100 | 70.72 | 94.97 | |
| CheXPO-v2-Phi4MM(1k)Model Category=Our Proposal, Optimization=Knowledge Graph Consistency GRPO, Sample Size=1k, Backbone=Phi-4MM2025.12 | 82.8 | 87.55 | 82.96 | 78.18 | 52 | 100 | 54.55 | 100 | 73.77 | 92.31 | 83.33 | 70.72 | 89.95 | |
| CheXPO-Phi3.5V(30k)Model Category=Our Proposal (Prior version), Optimization=Counterfactual DPO, Sample Size=30k, Backbone=Phi-3.5V2025.12 | 79.74 | 91.39 | 80.68 | 49.35 | 69.12 | 98.87 | 65.72 | 91.18 | 90.51 | 66.03 | 76.54 | 63.12 | 93.64 | |
| CheXPO-Phi3.5V(20k)Model Category=Our Proposal (Prior version), Optimization=Counterfactual DPO, Sample Size=20k, Backbone=Phi-3.5V2025.12 | 77.34 | 89.55 | 76.84 | 46.4 | 67.1 | 98.58 | 65.9 | 91.13 | 80.88 | 66.38 | 73.77 | 60.87 | 91.1 | |
| Phi4MM+CoTModel Category=Baseline Implementation, Optimization=Supervised Fine-Tuning (SFT), Answer Format=Chain-of-Thought (CoT)2025.12 | 75.17 | 79.25 | 71.11 | 70.91 | 44 | 98.87 | 63.64 | 100 | 72.13 | 76.92 | 83.33 | 66.67 | 80.16 | |
| Phi3.5V+CoTModel Category=Baseline Implementation, Optimization=Supervised Fine-Tuning (SFT), Answer Format=Chain-of-Thought (CoT)2025.12 | 73.2 | 86.26 | 71.46 | 43.86 | 62.16 | 98.3 | 64.74 | 91.18 | 78.8 | 63.62 | 70.68 | 57.53 | 87.77 | |
| ToolSelectType=Selection Method2026.02 | 72.01 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Phi4MM+SAModel Category=Baseline Implementation, Optimization=Supervised Fine-Tuning (SFT), Answer Format=Short Answer (SA)2025.12 | 70 | 76.23 | 64.44 | 60 | 36 | 95 | 63.64 | 100 | 67.21 | 53.85 | 100 | 59.46 | 76.19 | |
| Phi3.5V+SAModel Category=Baseline Implementation, Optimization=Supervised Fine-Tuning (SFT), Answer Format=Short Answer (SA)2025.12 | 69.95 | 81 | 62.89 | 44.87 | 60.03 | 95.69 | 63.84 | 91.18 | 73.21 | 63.45 | 66.98 | 55.82 | 81.76 | |
| EloType=Selection Method2026.02 | 63.74 | — | — | — | — | — | — | — | — | — | — | — | — | |
| NVIDIA CXR NV-Reason-3BShort name=NV, Type=Tool Candidate2026.02 | 63.22 | — | — | — | — | — | — | — | — | — | — | — | — | |
| HuatuoV-34BModel Category=Medical Visual Language Model, Parameters=34B2025.12 | 61.96 | 90.82 | 61.55 | 22.92 | 54.87 | 89.24 | 14.76 | 0 | 27.59 | 92.73 | 66.67 | 27.27 | 90.97 | |
| Qwen-30BShort name=Q30B, Type=Tool Candidate2026.02 | 60.63 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-OneVisionShort name=LOV, Type=Tool Candidate2026.02 | 58.91 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen-4BShort name=Q4B, Type=Tool Candidate2026.02 | 57.76 | — | — | — | — | — | — | — | — | — | — | — | — | |
| HuatuoV-7BModel Category=Medical Visual Language Model, Parameters=7B2025.12 | 57.16 | 76.42 | 63.88 | 22.36 | 32.96 | 83 | 20 | 48.08 | 24.83 | 83.64 | 53.33 | 26.17 | 83.07 | |
| Qwen-8BShort name=Q8B, Type=Tool Candidate2026.02 | 56.32 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-7BShort name=L7B, Type=Tool Candidate2026.02 | 54.6 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-13BShort name=L13B, Type=Tool Candidate2026.02 | 53.74 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2.5VL-Chat-7BModel Category=General Vision Language Model, Parameters=7B2025.12 | 53.36 | 66.95 | 63.02 | 21.67 | 18.81 | 88.95 | 14.29 | 81.25 | 21.38 | 45.45 | 36.67 | 25.91 | 76.31 | |
| Qwen-2BShort name=Q2B, Type=Tool Candidate2026.02 | 53.16 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-Med-7BModel Category=Medical Visual Language Model, Parameters=7B2025.12 | 47.6 | 68.21 | 64.22 | 17.92 | 0 | 68.56 | 27.14 | 0 | 16.55 | 45.45 | 70 | 11.51 | 77.78 | |
| CheXagent-8BShort name=CXA, Type=Tool Candidate2026.02 | 47.41 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SVMType=Selection Method2026.02 | 47.37 | — | — | — | — | — | — | — | — | — | — | — | — | |
| GraphType=Selection Method2026.02 | 47.37 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Causal LLMType=Selection Method2026.02 | 47.37 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-v1.6-7BModel Category=General Vision Language Model, Parameters=7B2025.12 | 47.26 | 67.67 | 63.97 | 17.92 | 0 | 67.71 | 23.81 | 0 | 15.17 | 58.18 | 70 | 11.07 | 77.52 | |
| MLPType=Selection Method2026.02 | 47.08 | — | — | — | — | — | — | — | — | — | — | — | — | |
| RandomType=Selection Method2026.02 | 46.49 | — | — | — | — | — | — | — | — | — | — | — | — | |
| RouterDCType=Selection Method2026.02 | 45.61 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Med-FlamingoModel Category=Medical Visual Language Model2025.12 | 43.64 | 62.45 | 57.67 | 17.36 | 14.16 | 50.99 | 12.38 | 0 | 17.93 | 65.45 | 50 | 14.84 | 67.72 | |
| Hybrid LLMType=Selection Method2026.02 | 42.69 | — | — | — | — | — | — | — | — | — | — | — | — | |
| KNNType=Selection Method2026.02 | 41.52 | — | — | — | — | — | — | — | — | — | — | — | — | |
| RaDialogShort name=RDlg, Type=Tool Candidate2026.02 | 38.51 | — | — | — | — | — | — | — | — | — | — | — | — | |
| RadFMModel Category=Medical Visual Language Model2025.12 | 38.24 | 59.51 | 48.53 | 12.36 | 7.08 | 40.79 | 8.57 | 0 | 13.79 | 63.64 | 63.33 | 9.75 | 62.06 | |
| Phi4MM-6BModel Category=General Vision Language Model, Parameters=6B2025.12 | 38.17 | 51.7 | 29.63 | 23.64 | 0 | 35 | 0 | 66.67 | 27.87 | 38.46 | 66.67 | 21.62 | 47.88 | |
| MFType=Selection Method2026.02 | 36.55 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaMA-3.2-VisionShort name=L3.2V, Type=Tool Candidate2026.02 | 33.05 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Phi3.5V-4.2BModel Category=General Vision Language Model, Parameters=4.2B2025.12 | 31.3 | 57.96 | 27.47 | 11.19 | 1.11 | 38.03 | 0 | 0 | 17.24 | 18.18 | 63.33 | 7.47 | 51.19 | |
| BLIP-CXRShort name=BCXR, Type=Tool Candidate2026.02 | 25 | — | — | — | — | — | — | — | — | — | — | — | — | |
| AutoMixType=Selection Method2026.02 | 24.27 | — | — | — | — | — | — | — | — | — | — | — | — | |
| BLIP-MIMICShort name=BMIM, Type=Tool Candidate2026.02 | 23.6 | — | — | — | — | — | — | — | — | — | — | — | — |