Resource Efficiency on ToolBench (Tool Use)
5.6Energy (Wh)Benign
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BenignTarget Model=Llama-3.32026.01 | 5.6 | 289.7 | 30 | |
| BenignTarget Model=Qwen-32026.01 | 6.1 | 240.3 | 30 | |
| BenignTarget Model=L-DeepSeek2026.01 | 6.8 | 334.7 | 40 | |
| BenignTarget Model=Mistral-L2026.01 | 7.3 | 368.1 | 40 | |
| BenignTarget Model=GLM-4.52026.01 | 7.4 | 327.5 | 20 | |
| OverthinkTarget Model=Llama-3.32026.01 | 11.9 | 331.3 | 50 | |
| BenignTarget Model=Seed-32B2026.01 | 15.2 | 301.2 | 80 | |
| OverthinkTarget Model=L-DeepSeek2026.01 | 25.9 | 334.6 | 130 | |
| OverthinkTarget Model=Mistral-L2026.01 | 48.3 | 387.1 | 320 | |
| OverthinkTarget Model=Seed-32B2026.01 | 48.3 | 311 | 510 | |
| OverthinkTarget Model=Qwen-32026.01 | 118.1 | 349.2 | 670 | |
| OverthinkTarget Model=GLM-4.52026.01 | 241.6 | 395.3 | 780 | |
| MCTS-based tool-layer DoS attackTarget Model=Qwen-32026.01 | 1,829.1 | 538.7 | 4,400 | |
| MCTS-based tool-layer DoS attackTarget Model=GLM-4.52026.01 | 2,401 | 447.6 | 3,670 | |
| MCTS-based tool-layer DoS attackTarget Model=L-DeepSeek2026.01 | 2,504.1 | 527.7 | 6,560 | |
| MCTS-based tool-layer DoS attackTarget Model=Seed-32B2026.01 | 2,779.9 | 588.3 | 5,170 | |
| MCTS-based tool-layer DoS attackTarget Model=Mistral-L2026.01 | 3,127.1 | 576.5 | 7,390 | |
| MCTS-based tool-layer DoS attackTarget Model=Llama-3.32026.01 | 3,159 | 572.2 | 6,650 |