Scientific Reasoning on GPQA Diamond (Acc, Len, CR)
63.1AccuracyASAG
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ASAGBackbone=Qwen3-14B2026.06 | 63.1 | 5,075 | 66 | |
| DEERBackbone=Qwen3-14B2026.06 | 60.6 | 4,998 | 65 | |
| DynasorBackbone=Qwen3-14B2026.06 | 59.6 | 5,387 | 70.1 | |
| TALEBackbone=Qwen3-14B2026.06 | 59.1 | 5,391 | 70.1 | |
| VanillaBackbone=Qwen3-14B2026.06 | 58.6 | 7,688 | 100 | |
| ASAGBackbone=Qwen3-8B2026.06 | 56.1 | 5,714 | 60.9 | |
| DynasorBackbone=Qwen3-8B2026.06 | 55.1 | 5,647 | 60.2 | |
| DEERBackbone=Qwen3-8B2026.06 | 54.5 | 5,334 | 56.9 | |
| VanillaBackbone=Qwen3-8B2026.06 | 52.5 | 9,382 | 100 | |
| TALEBackbone=Qwen3-8B2026.06 | 52 | 4,161 | 44.4 | |
| NoThinkingBackbone=Qwen3-14B2026.06 | 50.5 | 2,308 | 30 | |
| NoThinkingBackbone=Qwen3-8B2026.06 | 48.5 | 2,564 | 27.3 | |
| ASAGBackbone=Qwen3-4B2026.06 | 48 | 3,563 | 38.4 | |
| DEERBackbone=Qwen3-4B2026.06 | 47 | 3,802 | 41 | |
| VanillaBackbone=Qwen3-4B2026.06 | 46.5 | 9,277 | 100 | |
| DynasorBackbone=Qwen3-4B2026.06 | 46.5 | 4,179 | 45 | |
| TALEBackbone=Qwen3-4B2026.06 | 41.9 | 2,620 | 28.2 | |
| NoThinkingBackbone=Qwen3-4B2026.06 | 36.4 | 1,738 | 18.7 | |
| ASAGBackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 31.8 | 5,584 | 56.8 | |
| DynasorBackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 29.3 | 6,817 | 69.3 | |
| DEERBackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 28.8 | 5,639 | 57.3 | |
| NoThinkingBackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 26.8 | 1,385 | 14.1 | |
| TALEBackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 26.3 | 4,883 | 49.7 | |
| VanillaBackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 25.3 | 9,834 | 100 |