Token-level Hallucination Detection on AIME 2024 (AUROC/AUPRC)
87.39AUROCTOKENHD-8B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TOKENHD-8B2026.05 | 87.39 | 73.59 | |
| o4-mini2026.05 | 78.8 | 59.08 | |
| GPT-4.12026.05 | 72.06 | 44.26 | |
| QwQ-32B2026.05 | 69.04 | 49.19 | |
| R1-Qwen3-8B2026.05 | 64.84 | 45.62 |