Commonsense Reasoning on CommonSenseQA (Pass@1, #Tok)
85.5Pass@1 AccuracyDyCon
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DyConbackbone=QwQ-32B2026.06 | 85.5 | 584 | |
| Baselinebackbone=QwQ-32B2026.06 | 85.3 | 724 | |
| DyConbackbone=Qwen3-14B2026.06 | 83.9 | 946 | |
| Baselinebackbone=Qwen3-14B2026.06 | 83.6 | 1,038 | |
| DyConbackbone=Qwen3-4B-Thinking-25072026.06 | 79.4 | 2,190 | |
| Baselinebackbone=Qwen3-4B-Thinking-25072026.06 | 78.3 | 2,616 | |
| DyConbackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 65.6 | 540 | |
| Baselinebackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 64.7 | 746 |