Tool Reasoning on ToolBench (G1)
85.5Pass RateToolGate
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ToolGateModel=GPT-5.22026.01 | 85.5 | 83.5 | |
| ToolGateModel=Gemini 3 Pro2026.01 | 83 | 80.5 | |
| ToolChain*Model=GPT-5.22026.01 | 82.8 | 80 | |
| LATSModel=GPT-5.22026.01 | 80.3 | 78.8 | |
| ToolChain*Model=Gemini 3 Pro2026.01 | 80 | 78.5 | |
| LATSModel=Gemini 3 Pro2026.01 | 78.5 | 75.3 | |
| Tool-PlannerModel=GPT-5.22026.01 | 75.5 | 72.3 | |
| Tool-PlannerModel=Gemini 3 Pro2026.01 | 73.8 | 70 | |
| ToolGateModel=Deepseek V3.22026.01 | 72 | 70.3 | |
| DFSDTModel=GPT-5.22026.01 | 70 | 68.5 | |
| ToolChain*Model=Deepseek V3.22026.01 | 68.8 | 65 | |
| ToolGateModel=Qwen-3-235B2026.01 | 68.3 | 65.5 | |
| DFSDTModel=Gemini 3 Pro2026.01 | 68.3 | 65.5 | |
| LATSModel=Deepseek V3.22026.01 | 65.3 | 61.8 | |
| ToolChain*Model=Qwen-3-235B2026.01 | 65 | 62.8 | |
| ReACTModel=GPT-5.22026.01 | 63.5 | 62.8 | |
| LATSModel=Qwen-3-235B2026.01 | 62.5 | 59.3 | |
| Tool-PlannerModel=Deepseek V3.22026.01 | 62.5 | 60.3 | |
| Tool-PlannerModel=Qwen-3-235B2026.01 | 60.3 | 58 | |
| ReACTModel=Gemini 3 Pro2026.01 | 60 | 60.5 | |
| DFSDTModel=Deepseek V3.22026.01 | 58.5 | 55 | |
| DFSDTModel=Qwen-3-235B2026.01 | 57 | 53.8 | |
| ReACTModel=Deepseek V3.22026.01 | 52 | 48.5 | |
| ReACTModel=Qwen-3-235B2026.01 | 50.5 | — |