Visual Question Answering on PathVQA (Accuracy)
74.4AccuracySingle-Agent Best
Evaluation Results
| Method | Links | |
|---|---|---|
| Single-Agent BestBase Model=Gemma-3-4B, Image Resolution=3-5x reduced2025.08 | 74.4 | |
| Lingshu 7BMedical Samples=5M2026.06 | 71.96 | |
| MedGemma 27BMedical Samples=30M2026.06 | 67.2 | |
| MedAgentsBase Model=Gemma-3-4B, Image Resolution=3-5x reduced2025.08 | 66.9 | |
| ReConcileBase Model=Gemma-3-4B, Image Resolution=3-5x reduced2025.08 | 66.7 | |
| QoQ-Med-VL 7BMedical Samples=2.6M2026.06 | 64.12 | |
| Qwen2.5-VL-7B + OPENMEDREASONMedical Samples=450K2026.06 | 64.1 | |
| Qwen2.5-VL-7BMedical Samples=N/A2026.06 | 63.88 | |
| MedVL-Thinker 7BMedical Samples=200k2026.06 | 63.1 | |
| OctoMed 7BMedical Samples=8M2026.06 | 58.9 | |
| TMA-AllComponBase Model=Gemma-3-4B, Image Resolution=3-5x reduced2025.08 | 56.7 | |
| MDAgentsBase Model=Gemma-3-4B, Image Resolution=3-5x reduced2025.08 | 52 | |
| DyLanBase Model=Gemma-3-4B, Image Resolution=3-5x reduced2025.08 | 50 | |
| MedGemma 1.5 4BMedical Samples=30M2026.06 | 49.58 |