List your company

Step-level correctness assessment on Trips (test)

79.9PR-AUCReProbe
41.83651.71861.671.482Nov 9, 2025

Evaluation results

MethodLinks
ReProbe# Sample=32K, Architecture=Hidden States, Annotation=Self-anno2025.1179.9
ReProbe# Sample=32K, Architecture=Hidden States, Annotation=DeepSeek-anno2025.1179.3
ReProbe# Sample=32K, Architecture=Attn+Logit, Annotation=DeepSeek-anno2025.1174
ReProbe# Sample=32K, Architecture=Attn+Logit, Annotation=Self-anno2025.1173.5
Universal-PRM-Qwen2.5-Math-7B# Sample=690K2025.1173
Qwen2.5-Math-PRM-7B# Sample=860K2025.1171.1
Qwen2.5-Math-7B-PRM800k# Sample=265K2025.1170.8
Math-Shepherd-PRM-7B# Sample=440K2025.1166.2
H4-Qwen2.5-PRM-1.5B-0.2# Sample=369K2025.1159.7
CCP# Sample=-2025.1158.4
RLHFlow-PRM-Deepseek-8B# Sample=253K2025.1158.3
MaxProb# Sample=-2025.1157.8
Lexical Similarity# Sample=-2025.1156.9
Semantic Entropy# Sample=-2025.1156.5
MaxEntropy# Sample=-2025.1154.5
P(True)# Sample=-2025.1153.5
Degree Matrix# Sample=-2025.1153.4
Random# Sample=-2025.1152.4
RLHFlow-PRM-Mistral-8B# Sample=273K2025.1152.3
Perplexity# Sample=-2025.1151.9
Self-Certainty# Sample=-2025.1151.6
Skywork-PRM-1.5B# Sample=Unk2025.1143.3