Scientific Reasoning on GPQA D (Mean@16, Pass@16)
27.28Mean@16ADS
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ADSBackbone=Qwen2.5-Math-1.5B2026.06 | 27.28 | 89.39 | |
| DOTS+RRBackbone=Qwen2.5-Math-1.5B2026.06 | 20.78 | 77.61 | |
| ADSBackbone=Qwen3-4B-Base2026.06 | 20.69 | 71.55 | |
| GRPOBackbone=Qwen2.5-Math-1.5B2026.06 | 20.14 | 87.03 | |
| ADSBackbone=Olmo3-7B-SFT2026.06 | 17.84 | 69.36 | |
| GRPOBackbone=Olmo3-7B-SFT2026.06 | 16.56 | 69.53 | |
| GRPOBackbone=Qwen3-4B-Base2026.06 | 16.11 | 69.7 | |
| DOTS+RRBackbone=Olmo3-7B-SFT2026.06 | 13.34 | 63.47 | |
| BaseBackbone=Qwen2.5-Math-1.5B2026.06 | 12.58 | 80.13 | |
| BaseBackbone=Olmo3-7B-SFT2026.06 | 11.53 | 56.91 | |
| DOTS+RRBackbone=Qwen3-4B-Base2026.06 | 10.89 | 59.09 | |
| BaseBackbone=Qwen3-4B-Base2026.06 | 9.69 | 62.79 |