Share your thoughts, 1 month free Claude Pro on usSee more

Step-level Reasoning Verification on ProofNet

68.2PR-AUC

ReProbe, Hidden States, GPT-OSS-anno

Updated 1mo ago

Evaluation Results

Method	Links
ReProbe, Hidden States, GPT-OSS-anno 2025.11		68.2
Qwen2.5-Math-PRM-7B 2025.11		64.6
Universal-PRM-Qwen2.5-Math-7B 2025.11		56.8
Qwen2.5-Math-7B-PRM800K 2025.11		45.4
Skywork-PRM-1.5B 2025.11		28.9
Universal-PRM-Qwen2.5-Math-7B 2025.11		26.3
Qwen2.5-Math-7B 2025.11		24
Qwen2.5-Math-7B-PRM800k 2025.11		23.8
Math-Shepherd-PRM-7B 2025.11		18.8
Skywork-PRM-1.5B 2025.11		18.5
ReProbe, Attn+Logit, Qwen3-8B-anno 2025.11		15.5
H4-Qwen2.5-PRM-1.5B-0.2 2025.11		14.9
MaxProb 2025.11		12.3
RLHFlow-PRM-Deepseek-8B 2025.11		10.9
MaxEntropy 2025.11		10.7
H4-Qwen2.5-PRM-1.5B-0.2 2025.11		10.5
Perplexity 2025.11		9.9
RLHFlow-PRM-Mistral-8B 2025.11		9.3
Random 2025.11		8.2