Prompt Injection Defense on IHEval
88.1AccuracyAlways-DLLM
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Always-DLLMJudge=GPT-5.12026.05 | 88.1 | 5.2 | 81.4 | 1,540 | |
| SCOUTJudge=GPT-5.12026.05 | 86.1 | 4.2 | 76.4 | 1,242 | |
| SCOUTJudge=GPT-4o2026.05 | 84.4 | 13.8 | 82.6 | 524 | |
| Always-DLLMJudge=GPT-4o2026.05 | 80.9 | 0.8 | 62.6 | 1,476 |