Deception Detection on DeceptionBench (full 600-sample)
97.4Response AUROCDECOR
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DECORBackbone Model=Claude Opus 4.62026.05 | 97.4 | 96.7 | |
| DECORBackbone Model=Claude Sonnet 4.62026.05 | 96.4 | 97.3 | |
| DECORBackbone Model=Claude Opus 4.72026.05 | 96.4 | 96.4 | |
| DECORBackbone Model=GPT-5.42026.05 | 96 | 91.6 | |
| DECORBackbone Model=Gemini-3.1-pro-preview2026.05 | 94.7 | 84 | |
| DECORBackbone Model=GPT-4o2026.05 | 93.5 | 92 | |
| DECORBackbone Model=qwen3-235b-a22b2026.05 | 92.7 | 91.3 | |
| DECORBackbone Model=Gemini 2.5 Pro2026.05 | 92.6 | 90.1 | |
| DECORBackbone Model=DeepSeek-V3.22026.05 | 92.4 | 88.3 | |
| DECORBackbone Model=DeepSeek-R12026.05 | 91.6 | 86.9 | |
| DECORBackbone Model=GPT-52026.05 | 91.5 | 85.1 | |
| DECORBackbone Model=o32026.05 | 90.7 | 83 | |
| DECORBackbone Model=Grok-4.22026.05 | 90.1 | 77.9 | |
| DECORBackbone Model=o4-mini2026.05 | 89.4 | 85.5 | |
| DECORBackbone Model=DeepSeek-V3.12026.05 | 87.5 | 75.4 |