Open-ended reasoning on WildBench
57.05Creative ScorePPO w/ Structure Reward
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PPO w/ Structure RewardRL framework=PPO2026.03 | 57.05 | 45.87 | 27.7 | 53.47 | 30.05 | 40.26 | |
| GRPO w/ Structure RewardRL framework=GRPO2026.03 | 55.34 | 43.56 | 26.83 | 51.98 | 29.91 | 39.01 | |
| DPO2026.03 | 51.16 | 37.82 | 17.06 | 48.37 | 14.34 | 30.34 | |
| Base2026.03 | 51.01 | 36.23 | 16.35 | 48.71 | 14.72 | 29.91 | |
| GRPO w/ Entropy Minimization (EMPO)RL framework=GRPO2026.03 | 51.01 | 36.11 | 17.62 | 45.69 | 12.92 | 29.2 |