Mathematical Reasoning on AIME 2024 (Mean@32 accuracy)
75.2Mean@32 AccuracyHTPO
Evaluation Results
| Method | Links | |
|---|---|---|
| HTPOBackbone=Qwen3-8B-Instruct2026.05 | 75.2 | |
| SAPOBackbone=Qwen3-8B-Instruct2026.05 | 74 | |
| DAPOBackbone=Qwen3-8B-Instruct2026.05 | 73.2 | |
| GSPOBackbone=Qwen3-8B-Instruct2026.05 | 72.9 | |
| 80/20-RuleBackbone=Qwen3-8B-Instruct2026.05 | 72.9 | |
| BAPOBackbone=Qwen3-8B-Instruct2026.05 | 72.8 | |
| GRPO†Backbone=Qwen3-8B-Instruct2026.05 | 71.3 | |
| HTPOBackbone=Qwen3-8B-Base2026.05 | 41.5 | |
| SAPOBackbone=Qwen3-8B-Base2026.05 | 40.3 | |
| GSPOBackbone=Qwen3-8B-Base2026.05 | 39.2 | |
| BAPOBackbone=Qwen3-8B-Base2026.05 | 35.6 | |
| 80/20-RuleBackbone=Qwen3-8B-Base2026.05 | 34.3 | |
| DAPOBackbone=Qwen3-8B-Base2026.05 | 32.9 | |
| GRPO†Backbone=Qwen3-8B-Base2026.05 | 31.1 |