Pairwise Preference Ranking on UltraFeedback 10% holdout (test)
86.3Pairwise Accuracy (RM1-Honest)BENCHALIGN
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| BENCHALIGNModel set type=Arbitrary2026.02 | 86.3 | 86.2 | 0.894 | 0.891 | |
| RANDOMModel set type=Arbitrary2026.02 | 70.4 | 69.6 | 0.588 | 0.571 | |
| METABENCHModel set type=Arbitrary2026.02 | 70.2 | 69.8 | 0.58 | 0.566 | |
| TINYBENCHMARKSModel set type=Arbitrary2026.02 | 70.1 | 69.8 | 0.581 | 0.567 |