AI-Text Detection on AI-Generated Text (Held-out Set)
100AI-detection Rate (Aligned, Pangram)PASTA
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| PASTAModel=Gemma-2-9B2026.05 | 100 | -98 | -20.8 | -45.6 | -22.4 | |
| PASTAModel=Llama-3.1-8B-Inst.2026.05 | 99.8 | -73.2 | -26.4 | -24.4 | -13.4 | |
| PASTAModel=Qwen2.5-0.5B2026.05 | 99.2 | -9.8 | -3.8 | -29.4 | -10.7 | |
| PASTAModel=Qwen2.5-1.5B2026.05 | 99 | -89.7 | -10.8 | -87.2 | -18.9 | |
| PASTAModel=Mistral-7B-v0.32026.05 | 99 | -36.4 | -42.9 | -7 | -4.1 | |
| PASTAModel=Tulu-3-8B-RLVR2026.05 | 99 | -6 | 8.5 | -0.4 | -2.9 | |
| PASTAModel=Qwen2.5-14B2026.05 | 98.8 | -94.8 | -32.6 | -35.8 | -14.5 | |
| PASTAModel=Qwen2.5-7B2026.05 | 98.6 | -87.2 | -26 | -14.4 | -12.1 | |
| PASTAModel=Qwen2.5-32B2026.05 | 98.2 | -95.6 | -58.8 | -47.3 | -13.1 | |
| PASTAModel=Qwen2.5-3B2026.05 | 97.4 | -74.9 | -47 | -40.2 | -13.4 | |
| PASTAModel=OLMo-3-7B-Inst.2026.05 | 95.6 | -5.8 | 12.2 | -2.4 | -2.1 |