Mathematical Reasoning on AIME 2025 (Average@16)
29.2Average@16Ours (Shorten Prefix)
Evaluation Results
| Method | Links | |
|---|---|---|
| Ours (Shorten Prefix)Backbone=Qwen3-4B (non-think)2026.01 | 29.2 | |
| GRPOBackbone=Qwen3-4B (non-think)2026.01 | 26 | |
| RTSBackbone=Qwen3-4B (non-think)2026.01 | 22.3 | |
| MixedBackbone=Qwen3-4B (non-think)2026.01 | 19.6 | |
| Ours (Shorten Prefix)Backbone=Qwen3-8B-Base2026.01 | 18.3 | |
| BaseBackbone=Qwen3-4B (non-think)2026.01 | 18.1 | |
| GRPOBackbone=Qwen3-8B-Base2026.01 | 17.2 | |
| Ours (Shorten Prefix)Backbone=Qwen3-4B-Base2026.01 | 16.6 | |
| RTSBackbone=Qwen3-8B-Base2026.01 | 16.3 | |
| MixedBackbone=Qwen3-8B-Base2026.01 | 15 | |
| RTSBackbone=Qwen3-4B-Base2026.01 | 14 | |
| MixedBackbone=Qwen3-4B-Base2026.01 | 12.3 | |
| GRPOBackbone=Qwen3-4B-Base2026.01 | 11.9 | |
| BaseBackbone=Qwen3-4B-Base2026.01 | 5.2 | |
| BaseBackbone=Qwen3-8B-Base2026.01 | 3.3 |