Visual Tool Reasoning on VisualToolBench (test)
12.85Average@4Agent-KB
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Agent-KBModel=Qwen3-VL-235B, Setting=Agent-KB2026.03 | 12.85 | 20.09 | 3.88 | |
| w/ ToolsModel=Qwen3-VL-235B, Setting=w/ Tools2026.03 | 11.8 | 19.62 | 3.82 | |
| XSkillModel=Qwen3-VL-235B, Setting=XSkill2026.03 | 11.52 | 20.56 | 5.12 | |
| w/ ToolsModel=Qwen3-VL-32B, Setting=w/ Tools2026.03 | 11.09 | 18.69 | 2.65 | |
| Dynamic CheatSheetModel=Qwen3-VL-235B, Setting=Dynamic CheatSheet2026.03 | 10.86 | 19.16 | 4.09 | |
| No ToolsModel=Qwen3-VL-235B, Setting=No Tools2026.03 | 10.51 | 17.76 | — | |
| Agent-KBModel=Qwen3-VL-32B, Setting=Agent-KB2026.03 | 10.28 | 16.82 | 2.88 | |
| XSkillModel=Qwen3-VL-32B, Setting=XSkill2026.03 | 10.28 | 19.43 | 4.19 | |
| Agent Workflow MemoryModel=Qwen3-VL-32B, Setting=Agent Workflow Memory2026.03 | 10.05 | 17.76 | 2.73 | |
| Agent Workflow MemoryModel=Qwen3-VL-235B, Setting=Agent Workflow Memory2026.03 | 10.04 | 18.22 | 3.71 | |
| Dynamic CheatSheetModel=Qwen3-VL-32B, Setting=Dynamic CheatSheet2026.03 | 9.11 | 15.89 | 2.54 | |
| No ToolsModel=Qwen3-VL-32B, Setting=No Tools2026.03 | 7.94 | 12.62 | — |