Mathematical Reasoning on AIME24 (EM, #Tool)
36.7EM AccuracyToRL-GRPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ToRL-GRPOBase Model=Qwen2.5-Math-7B, Tool Use=Yes2026.06 | 36.7 | 2.1 | |
| OTC-GRPOBase Model=Qwen2.5-Math-7B, Tool Use=Yes2026.06 | 36.7 | 0.7 | |
| ParetoPOBase Model=Qwen2.5-Math-7B, Tool Use=Yes2026.06 | 36.7 | 0.8 | |
| ParetoPOBase Model=Qwen2.5-Math-1.5B, Tool Use=Yes2026.06 | 30 | 0.8 | |
| Qwen2.5-Math-7B-Ins-TIRBase Model=Qwen2.5-Math-7B, Tool Use=Yes2026.06 | 26.7 | 1.6 | |
| MO-GRPOBase Model=Qwen2.5-Math-7B, Tool Use=Yes2026.06 | 26.7 | 1.8 | |
| ToRL-GRPOBase Model=Qwen2.5-Math-1.5B, Tool Use=Yes2026.06 | 23.3 | 2.2 | |
| OTC-GRPOBase Model=Qwen2.5-Math-1.5B, Tool Use=Yes2026.06 | 20 | 1.1 | |
| MO-GRPOBase Model=Qwen2.5-Math-1.5B, Tool Use=Yes2026.06 | 16.7 | 1.8 | |
| Qwen2.5-Math-1.5B-Ins-TIRBase Model=Qwen2.5-Math-1.5B, Tool Use=Yes2026.06 | 13.3 | 1.1 | |
| Qwen2.5-Math-1.5B-InsBase Model=Qwen2.5-Math-1.5B, Tool Use=No2026.06 | 10 | — | |
| Qwen2.5-Math-7B-InsBase Model=Qwen2.5-Math-7B, Tool Use=No2026.06 | 10 | — |