Throughput Efficiency on ToolBench
4,602Throughput (tokens/s)Benign
Evaluation Results
| Method | Links | |
|---|---|---|
| BenignModel=Qwen-32026.01 | 4,602 | |
| OverthinkModel=Qwen-32026.01 | 4,550 | |
| OverthinkModel=Seed-32B2026.01 | 4,058 | |
| BenignModel=Seed-32B2026.01 | 4,001 | |
| BenignModel=L-DeepSeek2026.01 | 3,812 | |
| BenignModel=GLM-4.52026.01 | 3,753 | |
| OverthinkModel=Llama-3.32026.01 | 3,728 | |
| OverthinkModel=L-DeepSeek2026.01 | 3,711 | |
| BenignModel=Llama-3.32026.01 | 3,594 | |
| OverthinkModel=GLM-4.52026.01 | 3,189 | |
| BenignModel=Mistral-L2026.01 | 2,898 | |
| OverthinkModel=Mistral-L2026.01 | 2,724 | |
| MCTS-based tool-layer DoS attackModel=GLM-4.52026.01 | 2,324 | |
| MCTS-based tool-layer DoS attackModel=L-DeepSeek2026.01 | 2,106 | |
| MCTS-based tool-layer DoS attackModel=Qwen-32026.01 | 1,793 | |
| MCTS-based tool-layer DoS attackModel=Mistral-L2026.01 | 1,716 | |
| MCTS-based tool-layer DoS attackModel=Llama-3.32026.01 | 1,672 | |
| MCTS-based tool-layer DoS attackModel=Seed-32B2026.01 | 1,417 |