Mathematical Reasoning on OlymMath (avg@16 accuracy, pass@8)
25.8Avg@16 AccuracyTEMPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TEMPOBackbone=Qwen3-14B-Base2026.04 | 25.8 | 50.2 | |
| Zero-RL (PPO)Backbone=Qwen3-14B-Base2026.04 | 24.2 | 51.6 | |
| TEMPOBackbone=OLMO3-7B-Base2026.04 | 23.5 | 51.6 | |
| Zero-RL (PPO)Backbone=OLMO3-7B-Base2026.04 | 18.9 | 43.3 | |
| TTRLBackbone=OLMO3-7B-Base2026.04 | 18.9 | 33 | |
| EMPOBackbone=OLMO3-7B-Base2026.04 | 18.7 | 32.9 | |
| TEMPOBackbone=Qwen3-8B-Base2026.04 | 18.7 | 43.3 | |
| TTRLBackbone=Qwen3-14B-Base2026.04 | 18.3 | 31.7 | |
| EMPOBackbone=Qwen3-14B-Base2026.04 | 17.7 | 31.9 | |
| MiMo-Zero-RL-7BBackbone=MiMo-Zero-RL-7B2026.04 | 16.7 | 36.1 | |
| Zero-RL (PPO)Backbone=Qwen3-8B-Base2026.04 | 15 | 39.9 | |
| OLMO3.1-Zero-RL-7BBackbone=OLMO3.1-Zero-RL-7B2026.04 | 14.3 | 42.3 | |
| EMPOBackbone=Qwen3-8B-Base2026.04 | 13.1 | 26.7 | |
| TTRLBackbone=Qwen3-8B-Base2026.04 | 11.4 | 25.3 | |
| Oat-Zero-7BBackbone=Oat-Zero-7B2026.04 | 11.1 | 22.6 |