Scientific Reasoning on GPQA-D (Pass@1, #Tok)
68.2Pass@1 AccuracyDyCon
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DyConbackbone=Qwen3-4B-Thinking-25072026.06 | 68.2 | 6,205 | |
| Baselinebackbone=QwQ-32B2026.06 | 67.7 | 7,732 | |
| DyConbackbone=QwQ-32B2026.06 | 67.7 | 5,699 | |
| Baselinebackbone=Qwen3-4B-Thinking-25072026.06 | 66.2 | 9,210 | |
| DyConbackbone=Qwen3-14B2026.06 | 65.2 | 6,180 | |
| Baselinebackbone=Qwen3-14B2026.06 | 65.1 | 7,654 | |
| DyConbackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 47 | 5,304 | |
| Baselinebackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 38.4 | 7,518 |