Token-level Hallucination Detection on Math 500
9.86S_incorQwen3-0.6B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3-0.6BCategory=Backbone Models2026.05 | 9.86 | 98.94 | |
| Qwen3-1.7BCategory=Backbone Models2026.05 | 24 | 99.93 | |
| Qwen3-4BCategory=Backbone Models2026.05 | 42.23 | 99.91 | |
| R1-Qwen3-8BCategory=Critic Models2026.05 | 43.95 | 99.8 | |
| GPT-4.1Category=Critic Models2026.05 | 46.6 | 98.56 | |
| Qwen3-8BCategory=Backbone Models2026.05 | 49.45 | 99.84 | |
| QwQ-32BCategory=Critic Models2026.05 | 55.05 | 99.78 | |
| TOKENHD-0.6BCategory=Detector Models2026.05 | 63.64 | 94.83 | |
| TOKENHD-1.7BCategory=Detector Models2026.05 | 64.39 | 98.06 | |
| TOKENHD-8BCategory=Detector Models2026.05 | 64.55 | 98.71 | |
| TOKENHD-4BCategory=Detector Models2026.05 | 64.61 | 98.1 | |
| o4-miniCategory=Critic Models2026.05 | 71.39 | 99.93 | |
| o3Category=Labeler Models2026.05 | 82.03 | 99.86 | |
| GPT-5Category=Labeler Models2026.05 | 86.68 | 99.68 |