Hallucination Detection on PubMedQA (AUROC)
94.8AUROCLatentAudit
Evaluation Results
| Method | Links | |
|---|---|---|
| LatentAuditModel Backbone=Qwen-3 (8B)2026.04 | 94.8 | |
| LatentAuditModel Backbone=Llama-3 (8B)2026.04 | 94.2 | |
| LatentAuditModel Backbone=Qwen-2.5 (7B)2026.04 | 93.8 | |
| LatentAuditModel Backbone=Llama-2 (7B)2026.04 | 93.1 | |
| LatentAuditModel Backbone=Mistral (7B)2026.04 | 92.5 |