List your company

Self-confidence estimation on AIME hard OOD 23–25 (test)

12.9AccuracyOCB
8.0129.28110.5511.819Feb 10, 2026

Evaluation results

MethodLinks
OCBBackbone=Qwen2.5-7B-Instruct, Training steps=5122026.0212.90.9070.0960.074
Group MeanBackbone=Qwen2.5-7B-Instruct, Training steps=5122026.0212.10.9370.0940.058
RLCRBackbone=Qwen2.5-7B-Instruct, Training steps=5122026.0211.90.9320.1320.072
OCBModel=Qwen2.5-7B-Base, Training steps=5122026.029.50.9020.1260.064
RLCRModel=Qwen2.5-7B-Base, Training steps=5122026.029.40.9420.0920.052
Group MeanModel=Qwen2.5-7B-Base, Training steps=5122026.028.20.9110.0940.056