Token-level Hallucination Detection on AIME 2025 (AUROC/AUPRC)
89.47AUROCTOKENHD-8B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TOKENHD-8B2026.05 | 89.47 | 80.14 | |
| o4-mini2026.05 | 82.26 | 60.88 | |
| GPT-4.12026.05 | 75.43 | 49.87 | |
| QwQ-32B2026.05 | 70.82 | 52.12 | |
| R1-Qwen3-8B2026.05 | 64.35 | 46.63 |