Multiple-choice Question Answering on GPQA Diamond set (test)
82.29Accuracy (Overall)o3
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| o3Training Data=-2026.03 | 82.29 | — | — | — | |
| Principia-4BTraining Data=Principia Collec.2026.03 | 57.48 | — | — | — | |
| Principia-4B-ZeroTraining Data=Principia Collec.2026.03 | 43.56 | — | — | — | |
| Principia-7B-ZeroTraining Data=Principia Collec.2026.03 | 39.71 | — | — | — | |
| GLOREBackbone=Qwen2.5-7B-Instruct2025.03 | 36.36 | 46.51 | 25.81 | 42.11 | |
| SFTBackbone=Qwen2.5-7B-Instruct2025.03 | 35.35 | 44.19 | 23.66 | 52.63 | |
| Princpia-8B-ZeroTraining Data=Principia Collec.2026.03 | 33.62 | — | — | — | |
| RoTBackbone=Qwen2.5-7B-Instruct2025.03 | 33.33 | 41.86 | 23.66 | 42.11 | |
| Zero-shot CoTBackbone=Qwen2.5-7B-Instruct2025.03 | 30.81 | 38.37 | 22.58 | 36.84 | |
| MathNeuroBackbone=Qwen2.5-7B-Instruct2025.03 | 30.3 | 37.21 | 22.58 | 36.84 | |
| GLOREBackbone=Llama3.1-8B-Instruct2025.03 | 30.3 | 27.91 | 29.03 | 47.37 | |
| SFTBackbone=Llama3.1-8B-Instruct2025.03 | 29.29 | 26.88 | 27.96 | 47.37 | |
| BoostStepBackbone=Qwen2.5-7B-Instruct2025.03 | 28.28 | 34.88 | 23.66 | 21.05 | |
| Qwen2.5-7B-BaseTraining Data=-2026.03 | 27.4 | — | — | — | |
| Few-shot CoTBackbone=Qwen2.5-7B-Instruct2025.03 | 26.77 | 33.72 | 20.43 | 26.32 | |
| RoTBackbone=Llama3.1-8B-Instruct2025.03 | 25.76 | 24.42 | 23.66 | 42.11 | |
| MathNeuroBackbone=Llama3.1-8B-Instruct2025.03 | 24.75 | 22.09 | 22.58 | 47.37 | |
| BoostStepBackbone=Llama3.1-8B-Instruct2025.03 | 21.21 | 20.93 | 18.28 | 36.84 | |
| Zero-shot CoTBackbone=Llama3.1-8B-Instruct2025.03 | 20.2 | 19.77 | 16.13 | 42.11 | |
| Few-shot CoTBackbone=Llama3.1-8B-Instruct2025.03 | 20.2 | 22.09 | 15.05 | 36.84 |