Adversarial Robustness Evaluation on MMLU (TSR, ASR)
75.25Task Success Rate (TSR)Qwen-Plus
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen-PlusAttack Scenario=Group Collusive2026.05 | 75.25 | 20.75 | |
| Qwen-PlusAttack Scenario=Independent2026.05 | 73.75 | 21.5 | |
| Qwen-PlusAttack Scenario=Cooperative2026.05 | 72 | 24.25 | |
| GPT-5-NanoAttack Scenario=Independent2026.05 | 53.5 | 43.75 | |
| GPT-5-NanoAttack Scenario=Group Collusive2026.05 | 46.25 | 52 | |
| GPT-5-NanoAttack Scenario=Cooperative2026.05 | 39.25 | 59 | |
| GPT-3.5-TurboAttack Scenario=Independent2026.05 | 32 | 66 | |
| GPT-3.5-TurboAttack Scenario=Group Collusive2026.05 | 31.25 | 67.5 | |
| GPT-3.5-TurboAttack Scenario=Cooperative2026.05 | 29.5 | 69 |