List your company

Mathematical Reasoning on AIME 24 (Accuracy Avg@8)

82.1Accuracy (Avg@8)CERL full
71.59674.32377.0579.777May 22, 2026

Evaluation results

MethodLinks
CERL fullTraining=SFT+RL2026.0582.1
Correctness-only RLTraining=RL2026.0579.6
HaloTraining=SFT+RL2026.0579.5
Long-answer SFTTraining=SFT2026.0578.4
Fixed-ratio erasure + free answerTraining=None2026.0576.4
Base Qwen3-8BTraining=N/A2026.0575.9
Length-penalty RLTraining=RL2026.0575.1
TokenSkipTraining=SFT2026.0572