Explanation Quality Evaluation on PubMedQA (test)
39.7Reasoning Soundness LossPubMed Reasoner
Evaluation Results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| PubMed Reasoner2026.03 | 39.7 | 5.7 | 54.6 | 3.584 | 40.8 | 3.8 | 55.4 | 3.601 | 39.2 | 7.4 | 53.4 | 3.584 | 38.7 | 5.4 | 55.9 | 3.587 | |
| Geminiretrieval=w/o retrieval2026.03 | 54.6 | 5.7 | 39.7 | 3.416 | 55.4 | 3.8 | 40.8 | 3.421 | 53.4 | 7.4 | 39.2 | 3.438 | 55.9 | 5.4 | 38.7 | 3.424 |