Science Question Answering on GPQA Diamond (Accuracy Tracking & Repair)
75.3Initial AccuracyDISC
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DISCModel=gpt-5.2, K=32026.06 | 75.3 | 79.3 | 18 | 10 | 1.8 | 36.7 | |
| CoVeModel=gpt-5.22026.06 | 75.3 | 69.7 | 17 | 28 | 0.61 | 34.7 | |
| Self-RefineModel=gpt-5.22026.06 | 75.3 | 75.3 | 13 | 13 | 1 | 26.5 | |
| DISCModel=gpt-4.1-nano, K=32026.06 | 54 | 54 | 0 | 0 | — | 0 | |
| CoVeModel=gpt-4.1-nano2026.06 | 54 | 53 | 9 | 11 | 0.82 | 9.9 | |
| Self-RefineModel=gpt-4.1-nano2026.06 | 54 | 54.5 | 3 | 2 | 1.5 | 3.3 |