Model Ranking Prediction on Helpsteer 70B+ Models Holdout (test)
77.8Pairwise Acc (RM1)BENCHALIGN
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BENCHALIGNHoldout=70B+ Models2026.02 | 77.8 | 62 | 0.707 | 0.333 | |
| METABENCHHoldout=70B+ Models2026.02 | 56.4 | 48.4 | 0.2 | -0.028 | |
| TINYBENCHMARKSHoldout=70B+ Models2026.02 | 54.7 | 49.4 | 0.16 | -0.004 | |
| RANDOMHoldout=70B+ Models2026.02 | 53.8 | 46.8 | 0.087 | -0.106 |