Mathematical Reasoning on AIME 2025 (Average@12 accuracy)
72.5Average@12 AccuracyATESD
Evaluation Results
| Method | Links | |
|---|---|---|
| ATESDBackbone=Qwen3-8B2026.05 | 72.5 | |
| ATESDBackbone=Qwen3-4B2026.05 | 71.39 | |
| OPSDBackbone=Qwen3-8B2026.05 | 70.8 | |
| GRPOBackbone=Qwen3-8B2026.05 | 68.9 | |
| OPSDBackbone=Qwen3-4B2026.05 | 68.3 | |
| GRPOBackbone=Qwen3-4B2026.05 | 68.1 | |
| Base (Instruct)Backbone=Qwen3-4B2026.05 | 66.4 | |
| Base (Instruct)Backbone=Qwen3-8B2026.05 | 65.6 | |
| SFTBackbone=Qwen3-8B2026.05 | 64.2 | |
| SFTBackbone=Qwen3-4B2026.05 | 62.3 | |
| ATESDBackbone=Qwen3-1.7B2026.05 | 44.72 | |
| OPSDBackbone=Qwen3-1.7B2026.05 | 43.9 | |
| GRPOBackbone=Qwen3-1.7B2026.05 | 38.3 | |
| Base (Instruct)Backbone=Qwen3-1.7B2026.05 | 36.7 | |
| SFTBackbone=Qwen3-1.7B2026.05 | 36.3 |