Medical Diagnosis on DDxPlus
96.6SimilarityMedExAgent-8B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MedExAgent-8BParameters=8B, Condition=w/o noise injection2026.05 | 96.6 | 94.3 | 94.4 | |
| MedExAgent-8BParameters=8B, Condition=w/o tool-match reward2026.05 | 96.2 | 93 | 92.9 | |
| MedExAgent-8BParameters=8B, Condition=SFT-only2026.05 | 95.7 | 94.6 | 94.6 | |
| MedExAgent-8BParameters=8B, Condition=full2026.05 | 95.3 | 93.7 | 93.9 | |
| MedExAgent-8BParameters=8B, Condition=w/o cost penalty2026.05 | 84.6 | 83.1 | 83.1 | |
| Aloe-Beta-70BModel Family=Aloe-Beta, Parameters=70B2026.05 | 71.6 | 50.3 | 67 | |
| Qwen3-32BModel Family=Qwen3, Parameters=32B2026.05 | 68.5 | 46.6 | 62.6 | |
| HuatuoGPT-o1-70BModel Family=HuatuoGPT-o1, Parameters=70B2026.05 | 66.4 | 41 | 47 | |
| Qwen3-14BModel Family=Qwen3, Parameters=14B2026.05 | 65.9 | 39.9 | 53.8 | |
| Llama-3.1-8B-InstructModel Family=Llama-3.1, Parameters=8B, Variant=Instruct2026.05 | 64 | 38.3 | 60.5 | |
| Qwen3-8BModel Family=Qwen3, Parameters=8B2026.05 | 60.5 | 32.8 | 46 | |
| MedGemma-27B-text-itModel Family=MedGemma, Parameters=27B, Variant=text-it2026.05 | 49.8 | 24.1 | 32.5 |