General Reasoning on SuperGPQA
28.46Avg@kSFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SFTBackbone=Qwen2.5-7B-Instruct2026.05 | 28.46 | 53.83 | |
| RPSFTBackbone=Qwen2.5-7B-Instruct2026.05 | 27.84 | 54.57 | |
| IWBackbone=Qwen2.5-7B-Instruct2026.05 | 25.99 | 51.11 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.05 | 25.62 | 49.38 | |
| DFTBackbone=Qwen2.5-7B-Instruct2026.05 | 25.25 | 45.93 | |
| SFTBackbone=Llama-3.1-8B-Instruct2026.05 | 24.38 | 54.81 | |
| RPSFTBackbone=Llama-3.1-8B-Instruct2026.05 | 23.77 | 53.58 | |
| IWBackbone=Llama-3.1-8B-Instruct2026.05 | 22.28 | 49.14 | |
| RPSFTBackbone=Qwen2.5-3B-Instruct2026.05 | 21.67 | 48.89 | |
| BaseBackbone=Qwen2.5-3B-Instruct2026.05 | 21.36 | 49.63 | |
| BaseBackbone=Llama-3.1-8B-Instruct2026.05 | 19.88 | 48.15 | |
| SFTBackbone=Qwen2.5-3B-Instruct2026.05 | 19.81 | 45.68 | |
| IWBackbone=Qwen2.5-3B-Instruct2026.05 | 19.38 | 43.7 | |
| DFTBackbone=Qwen2.5-3B-Instruct2026.05 | 19.26 | 44.94 | |
| DFTBackbone=Llama-3.1-8B-Instruct2026.05 | 17.72 | 38.02 |