Explanation Quality Evaluation on MMLU-CK (test)
44Reasoning Soundness Loss (%)PubMed Reasoner
Evaluation Results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| PubMed Reasoner2026.03 | 44 | 10.8 | 45.2 | 3.699 | 25.2 | 11.7 | 64.1 | 3.595 | 34.4 | 20 | 45.6 | 3.732 | 35.3 | 15.9 | 48.8 | 3.712 | |
| Geminiretrieval=w/o retrieval2026.03 | 45.2 | 10.8 | 44 | 3.307 | 64.1 | 11.7 | 25.2 | 3.209 | 45.6 | 20 | 34.4 | 3.525 | 48.8 | 15.9 | 35.3 | 3.386 |