Multimodal Tool-use Reasoning on TRBench Mathematics
93.55Alg. AccuracyQwen3-VL-235B-thinking + UCT
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Qwen3-VL-235B-thinking + UCTCate.=Tool-based2026.02 | 93.55 | 91.78 | 87.4 | 94.74 | 97.14 | 90.3 | |
| Gemini-2.5-pro + UCTCate.=Tool-based2026.02 | 80.65 | 64.38 | 72.05 | 100 | 91.43 | 75.11 | |
| CREATORCate.=Tool-based2026.02 | 79.57 | 75.34 | 65.35 | 89.47 | 82.86 | 71.94 | |
| CRAFTCate.=Tool-based2026.02 | 74.19 | 76.71 | 68.9 | 94.74 | 85.71 | 73.42 | |
| VanillaCate.=Tool-based2026.02 | 58.06 | 72.6 | 54.33 | 57.89 | 40 | 56.96 | |
| Gemini-2.5-proCate.=Basic-COT2026.02 | 53.76 | 53.42 | 42.52 | 63.16 | 82.86 | 50.21 | |
| Qwen3-VL-235B-thinkingCate.=Basic-COT2026.02 | 40.86 | 45.21 | 64.96 | 73.68 | 57.14 | 56.96 |