List your company

Model Ranking Prediction on UltraFeedback 70B+ Models Holdout (test)

77.4Pairwise Acc (RM1_Honest)BENCHALIGN
55.24860.99966.7572.501Feb 2, 2026

Evaluation results

MethodLinks
BENCHALIGNHoldout=70B+ Models2026.0277.474.60.7060.659
RANDOMHoldout=70B+ Models2026.0262.160.50.280.237
METABENCHHoldout=70B+ Models2026.0258.157.50.2460.216
TINYBENCHMARKSHoldout=70B+ Models2026.0256.1570.1970.197