Mathematical Reasoning on AMC23 (Avg@16, Average)
89.22Avg@16BuPO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BuPOBackbone=Qwen3-8B2025.12 | 89.22 | 66.36 | |
| PPOBackbone=Qwen3-8B2025.12 | 87.03 | 58.41 | |
| RLOOBackbone=Qwen3-8B2025.12 | 86.41 | 63.36 | |
| GRPOBackbone=Qwen3-8B2025.12 | 85.94 | 64.23 | |
| Reinforce++Backbone=Qwen3-8B2025.12 | 82.66 | 60.41 | |
| BuPOBackbone=Qwen3-4B2025.12 | 81.09 | 58.51 | |
| RLOOBackbone=Qwen3-4B2025.12 | 77.66 | 54 | |
| PPOBackbone=Qwen3-4B2025.12 | 77.03 | 55.22 | |
| GRPOBackbone=Qwen3-4B2025.12 | 76.88 | 55.08 | |
| Verify-AlterBackbone=Qwen2.5-7B-Instruct2026.02 | 68 | — | |
| VanillaBackbone=Qwen3-4B2025.12 | 67.66 | 47.44 | |
| VanillaBackbone=Qwen3-8B2025.12 | 67.34 | 48.49 | |
| GenerateBackbone=Qwen2.5-7B-Instruct2026.02 | 65.3 | — | |
| Verify-InitBackbone=Qwen2.5-7B-Instruct2026.02 | 63.6 | — | |
| Reinforce++Backbone=Qwen3-4B2025.12 | 63.44 | 45.03 | |
| Mixed-TrainBackbone=Qwen2.5-7B-Instruct2026.02 | 59.2 | — | |
| GenerateBackbone=Qwen2.5-3B-Instruct2026.02 | 50.9 | — | |
| Mixed-TrainBackbone=Qwen2.5-3B-Instruct2026.02 | 49.4 | — | |
| Verify-InitBackbone=Qwen2.5-3B-Instruct2026.02 | 47.8 | — | |
| Verify-AlterBackbone=Qwen2.5-3B-Instruct2026.02 | 47.7 | — | |
| BuPOBackbone=Llama-OctoThinker-8B-Base2025.12 | 37.66 | 27.79 | |
| Verify-AlterBackbone=Qwen2.5-1.5B-Instruct2026.02 | 36.4 | — | |
| GRPOBackbone=Llama-OctoThinker-8B-Base2025.12 | 34.84 | 24.11 | |
| Reinforce++Backbone=Llama-OctoThinker-8B-Base2025.12 | 34.69 | 26.43 | |
| Mixed-TrainBackbone=Qwen2.5-1.5B-Instruct2026.02 | 33.3 | — | |
| Verify-InitBackbone=Qwen2.5-1.5B-Instruct2026.02 | 33 | — | |
| PPOBackbone=Llama-OctoThinker-8B-Base2025.12 | 31.72 | 22.82 | |
| GenerateBackbone=Qwen2.5-1.5B-Instruct2026.02 | 30.5 | — | |
| RLOOBackbone=Llama-OctoThinker-8B-Base2025.12 | 27.66 | 22.18 | |
| GRPOBackbone=Llama-OctoThinker-3B-Base2025.12 | 27.5 | 18.58 | |
| BuPOBackbone=Llama-OctoThinker-3B-Base2025.12 | 27.5 | 19.59 | |
| RLOOBackbone=Llama-OctoThinker-3B-Base2025.12 | 27.03 | 17.84 | |
| PPOBackbone=Llama-OctoThinker-3B-Base2025.12 | 22.19 | 16.69 | |
| Reinforce++Backbone=Llama-OctoThinker-3B-Base2025.12 | 9.38 | 5.27 | |
| VanillaBackbone=Llama-OctoThinker-8B-Base2025.12 | 4.53 | 3.75 | |
| VanillaBackbone=Llama-OctoThinker-3B-Base2025.12 | 1.24 | 1.68 |