Reverse Chain-of-Thought Generation on ArenaHard
72ScoreSSR-D
Evaluation Results
| Method | Links | |
|---|---|---|
| SSR-DBackbone=Qwen3-32B-Think2026.02 | 72 | |
| SSRBackbone=Qwen3-32B-Think2026.02 | 70.1 | |
| AUG-SUPBackbone=Qwen3-32B-Think2026.02 | 69.1 | |
| SUPBackbone=Qwen3-32B-Think2026.02 | 68.8 | |
| NEUBackbone=Qwen3-32B-Think2026.02 | 68.2 | |
| SSR-DBackbone=Qwen3-8B-Think2026.02 | 59.5 | |
| SSRBackbone=Qwen3-8B-Think2026.02 | 56.3 | |
| AUG-SUPBackbone=Qwen3-8B-Think2026.02 | 52.8 | |
| SUPBackbone=Qwen3-8B-Think2026.02 | 51.5 | |
| NEUBackbone=Qwen3-8B-Think2026.02 | 50.8 | |
| SSR-DBackbone=Qwen3-4B-Think-25072026.02 | 50.8 | |
| SSRBackbone=Qwen3-4B-Think-25072026.02 | 46.1 | |
| AUG-SUPBackbone=Qwen3-4B-Think-25072026.02 | 43.5 | |
| SUPBackbone=Qwen3-4B-Think-25072026.02 | 42.3 | |
| NEUBackbone=Qwen3-4B-Think-25072026.02 | 41.7 | |
| SSR-DBackbone=NBG4-3B-Base2026.02 | 37.2 | |
| SSRBackbone=NBG4-3B-Base2026.02 | 34.6 | |
| AUG-SUPBackbone=NBG4-3B-Base2026.02 | 33.2 | |
| SUPBackbone=NBG4-3B-Base2026.02 | 32.9 | |
| NEUBackbone=NBG4-3B-Base2026.02 | 32.4 |