Binary Disease Diagnosis on CheXpert (test)
85.9Macro AccCSRL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CSRLbackbone=Qwen2.5-VL-7B-Instruct, algorithm=GRPO2026.05 | 85.9 | 56.3 | |
| Best Tool per Pathologycategory=Tool Combination Baseline2026.05 | 80.1 | 51.6 | |
| CheXagent-8Bcategory=Specialist CXR Tool, parameters=8B2026.05 | 78.6 | 49.8 | |
| Logistic Regressioncategory=Tool Combination Baseline2026.05 | 78.4 | 51 | |
| Majority Voting (tie → neg)category=Tool Combination Baseline2026.05 | 76.6 | 49 | |
| CheXzerocategory=Specialist CXR Tool2026.05 | 76.3 | 48.9 | |
| InternVL3.5-38B-Instructcategory=Open-source General VLM, parameters=38B2026.05 | 74.4 | 45.1 | |
| Qwen2.5-VL-7B-Instructcategory=Open-source General VLM, parameters=7B2026.05 | 73.1 | 18.1 | |
| Hulu-Med-Flash-Preview-27Bcategory=Open-source Medical MLLM, parameters=27B2026.05 | 72.8 | 25.2 | |
| MedGemma-27B-ITcategory=Open-source Medical MLLM, parameters=27B2026.05 | 72.2 | 45.4 | |
| MedKLIPcategory=Specialist CXR Tool2026.05 | 71.3 | 46.4 | |
| HuatuoGPT-Vision-7Bcategory=Open-source Medical MLLM, parameters=7B2026.05 | 69.9 | 39.7 | |
| Majority Voting (tie → pos)category=Tool Combination Baseline2026.05 | 69.6 | 46.7 | |
| Gemini-3.1-Pro-Previewcategory=Closed-source General MLLM2026.05 | 68.5 | 43.6 | |
| DenseNet-CheXcategory=Specialist CXR Tool, Backbone=DenseNet-1212026.05 | 68.2 | 43.3 | |
| DenseNet-MIMICcategory=Specialist CXR Tool, Backbone=DenseNet-1212026.05 | 65.7 | 41.3 | |
| DenseNet-NIHcategory=Specialist CXR Tool, Backbone=DenseNet-1212026.05 | 60.7 | 37.9 | |
| GPT-4ocategory=Closed-source General MLLM2026.05 | 58.2 | 35.5 | |
| Lingshu-7Bcategory=Open-source Medical MLLM, parameters=7B2026.05 | 53.4 | 38.6 | |
| Claude-Sonnet-4.6category=Closed-source General MLLM2026.05 | 47.3 | 33.7 | |
| Qwen2.5-VL-72B-Instructcategory=Open-source General VLM, parameters=72B2026.05 | 38.2 | 30.9 |