Diversity of solution strategies on AIME 2024 + 2025
3.9Avg DiversityQwen2.5-7B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen2.5-7BMethod=Backbone2026.05 | 3.9 | 0.3 | 3.9 | 3.9 | |
| Qwen2.5-7B + DISAMethod=RL-post-trained variant2026.05 | 3.72 | 0.71 | 3.83 | 3.6 | |
| Qwen2.5-7B + FlowRLMethod=RL-post-trained variant2026.05 | 3.4 | 0.95 | 3.2 | 3.6 | |
| Qwen2.5-7B + GRPOMethod=RL-post-trained variant2026.05 | 3.02 | 0.72 | 2.93 | 3.1 | |
| Qwen2.5-7B + GSPOMethod=RL-post-trained variant2026.05 | 2.97 | 1.06 | 2.63 | 3.3 |