Mathematical Reasoning on AIME 2025 (avg@16 accuracy, pass@8)
44.6Avg@16 AccuracyEMPO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| EMPOBackbone=Qwen3-14B-Base2026.04 | 44.6 | 46.7 | |
| TEMPOBackbone=Qwen3-14B-Base2026.04 | 44.6 | 60 | |
| TEMPOBackbone=Qwen3-8B-Base2026.04 | 40.8 | 60.4 | |
| TTRLBackbone=Qwen3-14B-Base2026.04 | 40.8 | 45.8 | |
| Zero-RL (PPO)Backbone=Qwen3-14B-Base2026.04 | 37.1 | 59 | |
| TEMPOBackbone=OLMO3-7B-Base2026.04 | 37 | 52.5 | |
| EMPOBackbone=Qwen3-8B-Base2026.04 | 33.3 | 35.4 | |
| TTRLBackbone=Qwen3-8B-Base2026.04 | 32.8 | 33.3 | |
| MiMo-Zero-RL-7BBackbone=MiMo-Zero-RL-7B2026.04 | 32.3 | 51.9 | |
| TTRLBackbone=OLMO3-7B-Base2026.04 | 27.1 | 30.7 | |
| EMPOBackbone=OLMO3-7B-Base2026.04 | 26.7 | 29.5 | |
| OLMO3.1-Zero-RL-7BBackbone=OLMO3.1-Zero-RL-7B2026.04 | 26.5 | 39.7 | |
| Zero-RL (PPO)Backbone=OLMO3-7B-Base2026.04 | 26.3 | 41.1 | |
| Zero-RL (PPO)Backbone=Qwen3-8B-Base2026.04 | 25.4 | 44.8 | |
| Oat-Zero-7BBackbone=Oat-Zero-7B2026.04 | 12.3 | 33.7 |