Mathematical Reasoning on AIME 2024 (Average@16)
31.8Average@16Ours (Shorten Prefix)
Evaluation Results
| Method | Links | |
|---|---|---|
| Ours (Shorten Prefix)Backbone=Qwen3-4B (non-think)2026.01 | 31.8 | |
| RTSBackbone=Qwen3-4B (non-think)2026.01 | 31.2 | |
| GRPOBackbone=Qwen3-4B (non-think)2026.01 | 30.4 | |
| Ours (Shorten Prefix)Backbone=Qwen3-8B-Base2026.01 | 22.7 | |
| GRPOBackbone=Qwen3-8B-Base2026.01 | 22.3 | |
| MixedBackbone=Qwen3-4B (non-think)2026.01 | 20.6 | |
| BaseBackbone=Qwen3-4B (non-think)2026.01 | 19.5 | |
| Ours (Shorten Prefix)Backbone=Qwen3-4B-Base2026.01 | 19.4 | |
| GRPOBackbone=Qwen3-4B-Base2026.01 | 18.3 | |
| RTSBackbone=Qwen3-4B-Base2026.01 | 17.5 | |
| MixedBackbone=Qwen3-8B-Base2026.01 | 16.7 | |
| RTSBackbone=Qwen3-8B-Base2026.01 | 15 | |
| MixedBackbone=Qwen3-4B-Base2026.01 | 14.9 | |
| BaseBackbone=Qwen3-4B-Base2026.01 | 7.7 | |
| BaseBackbone=Qwen3-8B-Base2026.01 | 6.7 |