Reasoning on InfoBench
89.2Easy ScoreRL w/ PaTaRM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| RL w/ PaTaRMBase Model=Qwen3-14B2025.10 | 89.2 | 89.2 | 89.2 | |
| DPO (RLCF)Base Model=Qwen3-14B2025.10 | 88.7 | 86.5 | 87.2 | |
| GPT-4oModel=GPT-4o2025.10 | 87.9 | 87.6 | 87.1 | |
| SFTBase Model=Qwen3-14B2025.10 | 87.4 | 86 | 86.4 | |
| RL w/ SkyworkBase Model=Qwen3-14B2025.10 | 87.1 | 88.1 | 87.8 | |
| Qwen3-14BModel=Qwen3-14B2025.10 | 86.3 | 86.9 | 86.7 | |
| DPO (RLCF)Base Model=Qwen2.5-7B-Base2025.10 | 85.6 | 77.2 | 79.8 | |
| RL w/ PaTaRMBase Model=Qwen2.5-7B-Base2025.10 | 83.7 | 84.6 | 84.3 | |
| SFTBase Model=Qwen2.5-7B-Base2025.10 | 80.9 | 67.8 | 71.8 | |
| RL w/ SkyworkBase Model=Qwen2.5-7B-Base2025.10 | 77.1 | 73.6 | 78.7 | |
| Qwen2.5-7B-BaseModel=Qwen2.5-7B-Base2025.10 | 67.6 | 65.2 | 66.7 |