Multi-task Model Evaluation on AIME24, AIME25, HMMT25, HE+, MBPP+ Average
58.4Average AccuracyTeacher Model
Evaluation Results
| Method | Links | |
|---|---|---|
| Teacher ModelModel=Qwen3-4B-Instruct-2507, Teacher-Student Overlap=Large2026.06 | 58.4 | |
| IW-OPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 55.3 | |
| IW-OPD10Student Model=Qwen3-4B, Training Step=10, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 55.1 | |
| OPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 54.3 | |
| OPD10Student Model=Qwen3-4B, Training Step=10, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 54 | |
| IW-OPDStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 39.1 | |
| BaseStudent Model=Qwen3-4B, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 38.9 | |
| IW-OPD10Student Model=Qwen3-1.7B, Training Step=10, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 38.1 | |
| OPDStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 37.9 | |
| OPD10Student Model=Qwen3-1.7B, Training Step=10, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 36.8 | |
| BaseStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 28.7 | |
| IW-OPDStudent Model=Qwen3-0.6B, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 21.9 | |
| IW-OPD10Student Model=Qwen3-0.6B, Training Step=10, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 20.6 | |
| OPDStudent Model=Qwen3-0.6B, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 20 | |
| OPD10Student Model=Qwen3-0.6B, Training Step=10, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 18.6 | |
| BaseStudent Model=Qwen3-0.6B, Teacher Model=Qwen3-4B-Instruct-25072026.06 | 12.6 |