Model Ranking Prediction on UltraFeedback 30B+ Models Holdout (test)
77.3Pairwise Acc (RM1_Honest)BENCHALIGN
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| BENCHALIGNHoldout=30B+ Models2026.02 | 77.3 | 76.3 | 0.73 | 0.713 | |
| RANDOMHoldout=30B+ Models2026.02 | 65 | 61.8 | 0.407 | 0.317 | |
| TINYBENCHMARKSHoldout=30B+ Models2026.02 | 61.8 | 60.6 | 0.355 | 0.303 | |
| METABENCHHoldout=30B+ Models2026.02 | 61.3 | 60.4 | 0.346 | 0.302 |