Mathematical Reasoning on Minerva (Avg@16, Pass@16)
24.1Avg@16TAO-RL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TAO-RLBase Model=Qwen3-4B-Base2026.06 | 24.1 | 42.65 | |
| SimpleTIRBase Model=Qwen3-4B-Base2026.06 | 23.81 | 40.44 | |
| DOTS+RRBackbone=Qwen3-4B-Base2026.06 | 21.35 | 40.19 | |
| ADSBackbone=Qwen3-4B-Base2026.06 | 20.89 | 40.69 | |
| ADSBackbone=Olmo3-7B-SFT2026.06 | 18.58 | 37.5 | |
| GRPOBackbone=Qwen3-4B-Base2026.06 | 18.4 | 39.58 | |
| GRPOBackbone=Olmo3-7B-SFT2026.06 | 17.21 | 37.38 | |
| ADSBackbone=Qwen2.5-Math-1.5B2026.06 | 17.2 | 33.21 | |
| DOTS+RRBackbone=Olmo3-7B-SFT2026.06 | 17.16 | 34.68 | |
| BaseBackbone=Olmo3-7B-SFT2026.06 | 16.37 | 37.13 | |
| DOTS+RRBackbone=Qwen2.5-Math-1.5B2026.06 | 14.79 | 32.36 | |
| GRPOBackbone=Qwen2.5-Math-1.5B2026.06 | 13.63 | 31.86 | |
| TAO-RLBase Model=Qwen2.5-1.5B2026.06 | 10.55 | 29.04 | |
| BaseBackbone=Qwen3-4B-Base2026.06 | 10.27 | 35.05 | |
| SimpleTIRBase Model=Qwen2.5-1.5B2026.06 | 10.04 | 28.31 | |
| BaseBackbone=Qwen2.5-Math-1.5B2026.06 | 8.9 | 28.8 | |
| Base modelBase Model=Qwen3-4B-Base2026.06 | 8.87 | 33.09 | |
| Base modelBase Model=Qwen2.5-1.5B2026.06 | 0.76 | 8.09 |