Model Ranking Prediction on UltraFeedback 13B+ Models Holdout (test)
74.8Pairwise Accuracy (RM1_Honest)BENCHALIGN
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| BENCHALIGNHoldout=13B+ Models2026.02 | 74.8 | 74.9 | 0.686 | 0.691 | |
| RANDOMHoldout=13B+ Models2026.02 | 60.3 | 60 | 0.28 | 0.272 | |
| METABENCHHoldout=13B+ Models2026.02 | 58.6 | 57.9 | 0.248 | 0.229 | |
| TINYBENCHMARKSHoldout=13B+ Models2026.02 | 57.2 | 56.3 | 0.209 | 0.184 |