LLM-generated text detection on DetectRL
97.97AUROC (Multi-Domain)IRM
Evaluation Results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| IRMModel Family=Llama-3.2-1B2026.04 | 97.97 | 93.75 | 97.24 | 92.12 | 97.19 | 92.01 | 90.23 | 90.72 | 90.87 | 82.34 | 83.34 | 94.48 | 90.78 | 91.77 | — | |
| IRMBackbone=Llama-3.2-3B2026.04 | 97.31 | 92.51 | 97.25 | 92.15 | 97.21 | 91.69 | 91.68 | 90.67 | 91.37 | 78.59 | 80.97 | 93.85 | 88.85 | 91.08 | — | |
| ReMoDetect2026.04 | 91.99 | 85.95 | 90.41 | 83.17 | 91.22 | 82.09 | 79.08 | 78.36 | 80.22 | 88.53 | 85.87 | 91.63 | 87.65 | 85.86 | — | |
| IRMBackbone=Llama-3.1-8B2026.04 | 90.52 | 84.67 | 89.01 | 83.18 | 88.7 | 82.13 | 81.1 | 79.36 | 81.47 | 74.28 | 76.78 | 82.61 | 76.18 | 82.31 | — | |
| IRMModel Family=Gemma-2-2B2026.04 | 84.65 | 77.7 | 86.56 | 77.43 | 88.38 | 79.02 | 76.97 | 76.62 | 78.64 | 60.3 | 58.15 | 81.19 | 76.04 | 77.05 | — | |
| IRMBackbone=Qwen2-8B2026.04 | 74.42 | 58.02 | 73.08 | 59.05 | 73.7 | 62.13 | 55.77 | 57.7 | 61.15 | 42.54 | 41.77 | 69.61 | 58.49 | 60.57 | — | |
| RM-Deberta-v3-large-v22026.04 | 73.08 | 65.01 | 70.37 | 64.91 | 71.34 | 63.44 | 56.7 | 62.28 | 61.21 | 79.68 | 72.71 | 81.48 | 74.62 | 68.99 | — | |
| Skywork-Reward-V2Backbone=Llama-3.1-8B2026.04 | 71.33 | 62.97 | 70.67 | 66.15 | 68.23 | 65.85 | 59.06 | 65.59 | 63.23 | 45.41 | 61.78 | 39.32 | 23.98 | 58.74 | — | |
| Skywork-Reward-V2Backbone=Qwen2-8B2026.04 | 68.21 | 57.01 | 67.94 | 62.23 | 66.08 | 55.31 | 54.21 | 61.97 | 54.88 | 40.19 | 44.9 | 35.9 | 19.82 | 52.97 | — | |
| Skywork-Reward-V2Backbone=Llama-3.2-3B2026.04 | 63.48 | 48.47 | 63.33 | 57.78 | 60.32 | 41.35 | 44.5 | 56.99 | 32.58 | 34.77 | 49.65 | 30.52 | 6.91 | 45.43 | — | |
| GRM-gemma2-2B2026.04 | 51.07 | 31.2 | 50.57 | 45.32 | 50.69 | 59.97 | 27.53 | 41.58 | 56.3 | 45.71 | 53.97 | 63.69 | 60.37 | 49.07 | — | |
| GRM-Llama3.2-3B2026.04 | 45.47 | 21.27 | 46.15 | 22.85 | 48.56 | 30.8 | 19.12 | 19.63 | 29.15 | 46.81 | 49.99 | 60.71 | 54.64 | 38.09 | — | |
| BinocularsEvaluation Protocol=Zero-shot2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 64.8 | |
| Fast-DetectGPTEvaluation Protocol=Zero-shot2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 73 | |
| GLTREvaluation Protocol=Zero-shot2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 71.4 | |
| Human-as-OutlierEvaluation Protocol=Supervised2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 87.7 | |
| MELDEvaluation Protocol=Supervised2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 98.5 | |
| ModernBERT-DetectEvaluation Protocol=Supervised2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 87.9 | |
| RADAREvaluation Protocol=Supervised2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 76.5 | |
| RepreGuardEvaluation Protocol=Supervised2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 69.4 | |
| RoBERTa-ChatGPTEvaluation Protocol=Supervised2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 75.8 | |
| RoBERTa-Large (OpenAI)Evaluation Protocol=Supervised2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 63.1 |