Mathematical Problem Solving on AIME 2025 (Avg@8)
73.33Avg@8EvoTrainer
Evaluation Results
| Method | Links | |
|---|---|---|
| EvoTrainerConfiguration=Proposed Method2026.06 | 73.33 | |
| RAGEN v2 SNR FilteringConfiguration=Algorithmic Baseline2026.06 | 72.5 | |
| Human-engineered RLConfiguration=Reference2026.06 | 71.67 | |
| RAGEN v1 FilteringConfiguration=Algorithmic Baseline2026.06 | 70.83 | |
| AutoResearchConfiguration=Reference2026.06 | 70.42 | |
| GSPO + Clip-HigherConfiguration=Algorithmic Baseline2026.06 | 70 | |
| GRPO + Clip-HigherConfiguration=Algorithmic Baseline2026.06 | 69.58 | |
| Group-Variance FilteringConfiguration=Algorithmic Baseline2026.06 | 68.75 | |
| Base model (no RL)RL=None2026.06 | 67.5 |