Scientific Question Answering on SciKnowEval
68.5Accuracy (Biology)PMD
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PMDModel=Qwen3-8B2026.07 | 68.5 | 82.8 | 74.7 | 82.9 | 77.2 | — | |
| PMDModel=OLMo3-Instruct-7B2026.07 | 66.1 | 81.3 | 68.2 | 77.4 | 73.3 | — | |
| SDPOModel=Qwen3-8B2026.07 | 63.6 | 80.6 | 72.8 | 80.4 | 74.4 | — | |
| GRPOModel=Qwen3-8B2026.07 | 63.2 | 73.2 | 70.6 | 70.7 | 69.4 | — | |
| MOPDBackbone=Qwen3-8B2026.05 | 55.69 | 74.29 | 76.25 | 78.59 | — | — | |
| SDPOModel=OLMo3-Instruct-7B2026.07 | 54.1 | 79 | 66.1 | 76.8 | 69.5 | — | |
| SDPOBackbone=Qwen3-8B2026.05 | 50.6 | 62.91 | 67.36 | 72.34 | — | — | |
| GRPOBackbone=Qwen3-8B2026.05 | 47.32 | 64.24 | 64.12 | 72.09 | — | — | |
| GRPOModel=OLMo3-Instruct-7B2026.07 | 45.8 | 70.1 | 63.3 | 76.5 | 63.9 | — | |
| Base PolicyModel=Qwen3-8B2026.07 | 32.4 | 41.6 | 58.5 | 59.2 | 47.9 | — | |
| Qwen3-8BBackbone=Qwen3-8B2026.05 | 30.89 | 42.98 | 58.44 | 65.59 | — | — | |
| sGPOBackbone=Qwen3-4B-Instruct, Total FLOPs (ExaFLOPs)=1.9, Profiling FLOPs (ExaFLOPs)=0.42026.06 | 27.8 | 64.3 | 65.7 | 74 | 58 | 61.9 | |
| DAPOBackbone=Qwen3-4B-Instruct, Total FLOPs (ExaFLOPs)=5.2, Profiling FLOPs (ExaFLOPs)=0.02026.06 | 26.8 | 63.3 | 66.6 | 74.8 | 57.9 | 61.6 | |
| Base modelBackbone=Qwen3-4B-Instruct2026.06 | 25.5 | 36.8 | 53 | 64.9 | 45.1 | 44.2 | |
| Base PolicyModel=OLMo3-Instruct-7B2026.07 | 15.4 | 22.6 | 35.9 | 36.7 | 27.7 | — |