Tool Use on ToolSandBox (test)
69.88Overall Score235B-A22B-Instruct
Evaluation Results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 235B-A22B-InstructGroup=Open-source Baselines2026.06 | 69.88 | 62.46 | 62.68 | 63.85 | 59.48 | 62.68 | 57.63 | 90.98 | 73.84 | 73.44 | 74.98 | 70.75 | 74.85 | 70.99 | 74.73 | 74.79 | |
| Traj.-level TRUSTBase Model=Qwen3-8B-Base2026.06 | 68.28 | 68.66 | 59.7 | 63.1 | 56.93 | 61.58 | 52.04 | 90.7 | 73.01 | 76.41 | 74.94 | 69.16 | 66.74 | 68.77 | 71.26 | 71.24 | |
| AUQBase Model=Qwen3-8B-Thinking2026.06 | 64.08 | 56.46 | 56.68 | 57.85 | 53.48 | 56.68 | 51.63 | 87.98 | 67.84 | 67.44 | 68.98 | 64.75 | 68.85 | 64.99 | 68.73 | 68.79 | |
| 30B-A3B-InstructGroup=Open-source Baselines2026.06 | 61.57 | 62.08 | 60.07 | 62.88 | 54.09 | 62.34 | 57.24 | 58.45 | 64.68 | 62.12 | 62.25 | 61.72 | 63.61 | 64.94 | 63.54 | 63.55 | |
| Traj.-level CM2Base Model=Qwen3-8B-Base2026.06 | 61.21 | 70.57 | 58.55 | 64.22 | 51.93 | 56.64 | 53.7 | 55.6 | 60.12 | 65.04 | 65.02 | 65.03 | 61.8 | 61.3 | 64.14 | 64.51 | |
| 8B-ThinkingGroup=Open-source Baselines2026.06 | 59.74 | 64.32 | 44.79 | 50.16 | 44.06 | 61.55 | 35.76 | 94.93 | 59.05 | 62.95 | 65.13 | 60.74 | 62.35 | 61.08 | 65.23 | 64 | |
| SAGEBase Model=Qwen3-8B-Thinking2026.06 | 59.72 | 54.85 | 51.56 | 53.95 | 47.56 | 59.41 | 43.68 | 83.48 | 59.7 | 65.72 | 60.27 | 58.29 | 62.97 | 66.61 | 62.22 | 65.52 | |
| Turn-level TRUSTBase Model=Qwen3-4B-Thinking2026.06 | 56.35 | 51.18 | 46.26 | 52.83 | 37.68 | 55.59 | 42.27 | 94.44 | 57.42 | 58.53 | 59.05 | 58.68 | 57.66 | 56.9 | 58.02 | 58.77 | |
| AUQBase Model=Qwen3-4B-Thinking2026.06 | 55.47 | 48.23 | 47.92 | 49.2 | 44.79 | 51.59 | 41.71 | 84.87 | 56.12 | 58.09 | 60.55 | 56.98 | 60.41 | 59.03 | 55.51 | 56.99 | |
| 4B-ThinkingGroup=Open-source Baselines2026.06 | 52.89 | 57.68 | 38.09 | 43.15 | 38.18 | 53.03 | 28.71 | 90.82 | 58.71 | 54.75 | 53.21 | 52.61 | 53.82 | 53.64 | 56.4 | 60.54 | |
| SAGEBase Model=Qwen3-4B-Thinking2026.06 | 52.54 | 48.22 | 46.93 | 49.14 | 42.04 | 54.38 | 40.99 | 71.06 | 50.14 | 56.03 | 54.33 | 51.6 | 60.46 | 51.49 | 55.9 | 55.45 | |
| Turn-level GRPOBase Model=Qwen3-4B-Thinking2026.06 | 39.37 | 46.92 | 21.15 | 28.67 | 19.04 | 23.37 | 19.05 | 95.98 | 43.36 | — | 40.93 | 38.86 | 44.28 | 37.9 | 41.82 | 43.71 |