Multimodal Tool-use Reasoning on TRBench Science
88Accuracy (Biology)Gemini-2.5-pro + UCT
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Gemini-2.5-pro + UCTCate.=Tool-based2026.02 | 88 | 71.03 | 85.19 | 79.17 | |
| Qwen3-VL-235B-thinking + UCTCate.=Tool-based2026.02 | 88 | 83.18 | 90.74 | 87.08 | |
| Gemini-2.5-proCate.=Basic-COT2026.02 | 84 | 43.93 | 37.96 | 45.42 | |
| Qwen3-VL-235B-thinkingCate.=Basic-COT2026.02 | 84 | 75.7 | 80.56 | 78.75 | |
| CRAFTCate.=Tool-based2026.02 | 80 | 71.96 | 81.48 | 77.08 | |
| CREATORCate.=Tool-based2026.02 | 76 | 76.64 | 80.56 | 78.34 | |
| VanillaCate.=Tool-based2026.02 | 48 | 29.91 | 26.85 | 30.42 |