Mathematical Reasoning on (AIME24, AIME25, AMC23, GSM8K, MATH500)
25.9AIME 2024 ScoreQwen2.5-7B + SFT-then-RL
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen2.5-7B + SFT-then-RLBackbone=Qwen2.5-7B, Training Protocol=SFT-then-RL2026.04 | 25.9 | 23.3 | 59.4 | 77.4 | 91.2 | |
| Qwen2.5-7B + E3-TIR (Ours)Backbone=Qwen2.5-7B, Training Protocol=E3-TIR2026.04 | 23.2 | 29.3 | 64.8 | 80.4 | 92.1 | |
| Qwen2.5-7B + Zero-RLBackbone=Qwen2.5-7B, Training Protocol=Zero-RL2026.04 | 21.3 | 27.1 | 62.1 | 76.8 | 90.6 | |
| Qwen2.5-3B + E3-TIR (Ours)Backbone=Qwen2.5-3B, Training Protocol=E3-TIR2026.04 | 19 | 20.8 | 56 | 72.3 | 85.2 | |
| Llama3.1-8B + Zero-RLBackbone=Llama3.1-8B, Training Protocol=Zero-RL2026.04 | 17.2 | 12.5 | 53.7 | 62.5 | 87.9 | |
| Llama3.1-8B + SFT-then-RLBackbone=Llama3.1-8B, Training Protocol=SFT-then-RL2026.04 | 16 | 12.5 | 52 | 64 | 86.1 | |
| Llama3.1-8B + E3-TIR (Ours)Backbone=Llama3.1-8B, Training Protocol=E3-TIR2026.04 | 15.5 | 20 | 58.4 | 65.2 | 88.2 | |
| Qwen2.5-7B + Only SFTBackbone=Qwen2.5-7B, Training Protocol=Only SFT2026.04 | 14.6 | 14.1 | 56.2 | 71.3 | 89.6 | |
| Qwen2.5-3B + Zero-RLBackbone=Qwen2.5-3B, Training Protocol=Zero-RL2026.04 | 12 | 13.3 | 48.7 | 71.9 | 85.4 | |
| Qwen2.5-3B + SFT-then-RLBackbone=Qwen2.5-3B, Training Protocol=SFT-then-RL2026.04 | 11.2 | 16.5 | 51.8 | 69.1 | 85.1 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B, Training Protocol=Instruct2026.04 | 7.5 | 9.2 | 50 | 74.1 | 89.4 | |
| Llama3.1-8B + Only SFTBackbone=Llama3.1-8B, Training Protocol=Only SFT2026.04 | 7.5 | 11.2 | 34.3 | 59.3 | 85.6 | |
| Qwen2.5-3B + Only SFTBackbone=Qwen2.5-3B, Training Protocol=Only SFT2026.04 | 6 | 10.8 | 35.6 | 58.3 | 81 | |
| Qwen2.5-3B-InstructBackbone=Qwen2.5-3B, Training Protocol=Instruct2026.04 | 4.5 | 8.6 | 38.3 | 65.3 | 81 | |
| Llama3.1-8B-InstructBackbone=Llama3.1-8B, Training Protocol=Instruct2026.04 | 0 | 1.6 | 25 | 46.6 | 78.3 |