Math Reasoning on OpenMathReasoning
14.2AccuracyCoScale-RL
Evaluation results
| Method | Links | |
|---|---|---|
| CoScale-RL2026.01 | 14.2 | |
| Scale-RL2026.01 | 5.8 | |
| SpeedControl2026.01 | 5.8 | |
| SAPO2026.01 | 5.7 | |
| GRPO2026.01 | 5.6 | |
| COMPASS2026.01 | 5.4 | |
| SFTed (w/ Data Scaling)Data Scaling=Yes2026.01 | 4.3 | |
| Pretrain + RL2026.01 | 3.5 | |
| Qwen2.5-0.5B Instruct2026.01 | 2.6 | |
| SFTed (w/o Data Scaling)Data Scaling=No2026.01 | 2.5 |