Ranking Consistency Analysis on MMLU-Pro health Human aging
0.62Spearman CorrelationMedQA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MedQAConstruction Method=Domain-specific comparison2026.05 | 0.62 | 0.42 | |
| SCICUSTOMConstruction Method=Grounded data and ontology-driven2026.05 | 0.49 | 0.37 | |
| MMLUConstruction Method=Domain-specific comparison2026.05 | 0.21 | 0.14 | |
| EmbeddingConstruction Method=Embedding-based k-NN retrieval2026.05 | 0.21 | 0.1 | |
| GPT-5Construction Method=Synthetic benchmark generation2026.05 | 0.2 | 0.18 | |
| GPQAConstruction Method=General scientific benchmark2026.05 | -0.1 | -0.1 | |
| IfBenchConstruction Method=General-purpose benchmark2026.05 | -0.32 | -0.14 | |
| SimpleQAConstruction Method=General-purpose benchmark2026.05 | -0.5 | -0.33 |