Mathematical Reasoning on AIME24 (Pass@1 Avg@32)
32.2Pass@1 AccuracyMix-RL
Evaluation Results
| Method | Links | |
|---|---|---|
| Mix-RLModel Backbone=Qwen2.5-Math-7B, Gen Mode=Fold2026.02 | 32.2 | |
| Unfold-RLModel Backbone=Qwen2.5-Math-7B, Gen Mode=Unfold2026.02 | 32 | |
| Fold-RLModel Backbone=Qwen2.5-Math-7B, Gen Mode=Fold2026.02 | 31.3 | |
| Unfold-RLModel Backbone=Qwen2.5-Math-7B, Gen Mode=Fold2026.02 | 29.1 | |
| Fold-RLModel Backbone=Qwen3-4B-Base, Gen Mode=Fold2026.02 | 28.4 | |
| Mix-RLModel Backbone=Qwen3-4B-Base, Gen Mode=Fold2026.02 | 27.6 | |
| Unfold-RLModel Backbone=Qwen3-4B-Base, Gen Mode=Unfold2026.02 | 27.5 | |
| Cold-StartModel Backbone=Qwen2.5-Math-7B, Gen Mode=Unfold2026.02 | 26.7 | |
| Zero-RLModel Backbone=Qwen2.5-Math-7B, Gen Mode=Unfold2026.02 | 25.8 | |
| Unfold-RLModel Backbone=Qwen3-4B-Base, Gen Mode=Fold2026.02 | 25.8 | |
| Zero-RLModel Backbone=Qwen3-4B-Base, Gen Mode=Unfold2026.02 | 25.5 | |
| Cold-StartModel Backbone=Qwen3-4B-Base, Gen Mode=Unfold2026.02 | 23.8 | |
| Cold-StartModel Backbone=Qwen2.5-Math-7B, Gen Mode=Fold2026.02 | 23 | |
| Cold-StartModel Backbone=Qwen3-4B-Base, Gen Mode=Fold2026.02 | 19.2 |