Ranking Consistency Analysis on MMLU-Pro health Virology
0.55Spearman CorrelationSCICUSTOM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SCICUSTOMConstruction Method=Grounded data and ontology-driven2026.05 | 0.55 | 0.45 | |
| MedQAConstruction Method=Domain-specific comparison2026.05 | 0.44 | 0.35 | |
| SimpleQAConstruction Method=General-purpose benchmark2026.05 | 0.35 | 0.31 | |
| MMLUConstruction Method=Domain-specific comparison2026.05 | 0.35 | 0.31 | |
| GPT-5Construction Method=Synthetic benchmark generation2026.05 | 0.25 | 0.2 | |
| EmbeddingConstruction Method=Embedding-based k-NN retrieval2026.05 | 0.18 | 0.11 | |
| IfBenchConstruction Method=General-purpose benchmark2026.05 | 0.04 | -0.05 | |
| GPQAConstruction Method=General scientific benchmark2026.05 | -0.11 | -0.05 |