Model Ranking Prediction on UltraFeedback 70B+ Models Holdout (test)
77.4Pairwise Acc (RM1_Honest)BENCHALIGN
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BENCHALIGNHoldout=70B+ Models2026.02 | 77.4 | 74.6 | 0.706 | 0.659 | |
| RANDOMHoldout=70B+ Models2026.02 | 62.1 | 60.5 | 0.28 | 0.237 | |
| METABENCHHoldout=70B+ Models2026.02 | 58.1 | 57.5 | 0.246 | 0.216 | |
| TINYBENCHMARKSHoldout=70B+ Models2026.02 | 56.1 | 57 | 0.197 | 0.197 |