Medical Visual Question Answering and Reasoning on CheXbench
76.5Rad-Restruct AccRadAgents
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RadAgentsModel / Backbone=RadAgents, V-RAG=true (k=3)2025.09 | 76.5 | 89.4 | 69.2 | 74.6 | |
| Qwen3-VL w/ WorkflowModel / Backbone=Qwen3-VL, Agent Workflow Strategy=Workflow templates2025.09 | 72.2 | 87.8 | 65.3 | 71 | |
| RadAgents wo/ V-RAGModel / Backbone=RadAgents, V-RAG=false2025.09 | 71.3 | 87.8 | 66.3 | 71.5 | |
| Qwen3-VL w/ ReActModel / Backbone=Qwen3-VL, Agent Workflow Strategy=ReAct2025.09 | 70.4 | 86.2 | 61.3 | 68 | |
| CheXagentModel / Backbone=CheXagent2025.09 | 57.1 | 78.1 | 59 | 64.7 | |
| GPT-4oModel / Backbone=GPT-4o2025.09 | 53.9 | 85.4 | 51.1 | 63.5 |