Multi-hop QA Reasoning on MuSiQue (Degen, Len, AES)
0DegenBase
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BaseModel Scale=Qwen3-0.6B2026.02 | 0 | 731 | 0 | |
| PromptModel Scale=Qwen3-0.6B2026.02 | 0 | 697 | -0.23 | |
| MBT-SModel Scale=Qwen3-4B2026.02 | 0 | 471 | 0.85 | |
| BaseModel Scale=Qwen3-1.7B2026.02 | 1 | 1,186 | 0 | |
| GRPOModel Scale=Qwen3-1.7B2026.02 | 1 | 1,379 | 0.56 | |
| MBT-RModel Scale=Qwen3-1.7B2026.02 | 1 | 728 | 1.06 | |
| MBT-RModel Scale=Qwen3-4B2026.02 | 1 | 724 | 0.65 | |
| PromptModel Scale=Qwen3-1.7B2026.02 | 2 | 1,134 | 0.05 | |
| MBT-SModel Scale=Qwen3-1.7B2026.02 | 2 | 521 | 1.37 | |
| BaseModel Scale=Qwen3-4B2026.02 | 2 | 1,368 | 0 | |
| MBT-RModel Scale=Qwen3-0.6B2026.02 | 5 | 750 | 2.21 | |
| PromptModel Scale=Qwen3-4B2026.02 | 5 | 1,324 | -0.06 | |
| MBT-SModel Scale=Qwen3-0.6B2026.02 | 6 | 561 | 2.72 | |
| RSModel Scale=Qwen3-0.6B2026.02 | 12 | 943 | -0.04 | |
| GRPOModel Scale=Qwen3-0.6B2026.02 | 16 | 2,167 | -0.43 | |
| RSModel Scale=Qwen3-1.7B2026.02 | 19 | 1,533 | 0.03 | |
| GRPOModel Scale=Qwen3-4B2026.02 | 73 | 2,433 | -0.65 | |
| RSModel Scale=Qwen3-4B2026.02 | 133 | 3,097 | -1.36 | |
| TokenSkipModel Scale=Qwen3-0.6B2026.02 | 174 | 3,220 | -3.7 | |
| LIMOProModel Scale=Qwen3-0.6B2026.02 | 386 | 5,697 | -5.26 | |
| TokenSkipModel Scale=Qwen3-1.7B2026.02 | 471 | 7,437 | -5.66 | |
| LIMOProModel Scale=Qwen3-1.7B2026.02 | 585 | 8,393 | -5.9 | |
| LIMOProModel Scale=Qwen3-4B2026.02 | 590 | 8,478 | -6.13 | |
| TokenSkipModel Scale=Qwen3-4B2026.02 | 651 | 9,932 | -7.69 |