Search-based Question Answering on SQA CS V2 (test)
57.62IRCo-ReAct
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Co-ReActSearch Agent=Qwen3-14B, Answer Rewriter=Qwen3-235B2026.05 | 57.62 | 75.29 | 73.82 | 97.48 | 76.05 | |
| Co-ReActSearch Agent=Qwen3-8B, Answer Rewriter=Qwen3-235B2026.05 | 50.91 | 79.99 | 73.45 | 94.84 | 74.8 | |
| CRITICSearch Agent=Qwen3-14B, Answer Rewriter=Qwen3-235B2026.05 | 50.63 | 75.96 | 71.63 | 95.36 | 73.4 | |
| CRITICSearch Agent=Qwen3-8B, Answer Rewriter=Qwen3-235B2026.05 | 50.56 | 74.93 | 68.77 | 92.5 | 71.69 | |
| Self-RefineSearch Agent=Qwen3-8B, Answer Rewriter=Qwen3-235B2026.05 | 50.25 | 77.07 | 72.71 | 96.7 | 74.18 | |
| Self-RefineSearch Agent=Qwen3-14B, Answer Rewriter=Qwen3-235B2026.05 | 49.72 | 77.69 | 71.81 | 97.65 | 74.22 | |
| ReActSearch Agent=Qwen3-8B, Answer Rewriter=Qwen3-235B2026.05 | 49.01 | 76.74 | 71.29 | 93.98 | 72.76 | |
| ReActSearch Agent=Qwen3-14B, Answer Rewriter=Qwen3-235B2026.05 | 48.23 | 73.43 | 71.58 | 97.67 | 72.73 | |
| Best-of-NSearch Agent=Qwen3-8B, Answer Rewriter=Qwen3-235B2026.05 | 45.77 | 75.66 | 67.93 | 90.98 | 70.08 | |
| Best-of-NSearch Agent=Qwen3-14B, Answer Rewriter=Qwen3-235B2026.05 | 44.98 | 78.4 | 69.25 | 90.61 | 70.81 | |
| Step-BackSearch Agent=Qwen3-14B, Answer Rewriter=Qwen3-235B2026.05 | 43.61 | 82.19 | 62.31 | 84.65 | 68.19 | |
| Step-BackSearch Agent=Qwen3-8B, Answer Rewriter=Qwen3-235B2026.05 | 43.02 | 81.08 | 66.24 | 85.65 | 69 |