Hallucination Detection on Average (CoQA, SQuAD v2.0, TriviaQA, TruthfulQA)
92AUROCPCNET
Evaluation results
| Method | Links | ||
|---|---|---|---|
| PCNETModel=Mistral-7B-v0.32026.05 | 92 | 91 | |
| PCNETModel=Qwen3-4B2026.05 | 92 | 91 | |
| PCNETModel=Llama-3.1-8B2026.05 | 92 | 91 | |
| PCNETModel=Llama-3.2-1B2026.05 | 88 | 88 | |
| HaloScopeModel=Llama-3.1-8B2026.05 | 87 | 88 | |
| HaloScopeModel=Llama-3.2-1B2026.05 | 82 | 81 | |
| HaloScopeModel=Mistral-7B-v0.32026.05 | 79 | 82 | |
| HaloScopeModel=Qwen3-4B2026.05 | 67 | 73 | |
| Token NLLModel=Mistral-7B-v0.32026.05 | 63 | 70 | |
| SEPModel=Llama-3.2-1B2026.05 | 62 | 70 | |
| SEPModel=Mistral-7B-v0.32026.05 | 62 | 69 | |
| Token NLLModel=Llama-3.1-8B2026.05 | 62 | 69 | |
| Token NLLModel=Qwen3-4B2026.05 | 61 | 69 | |
| SEPModel=Qwen3-4B2026.05 | 60 | 70 | |
| Token NLLModel=Llama-3.2-1B2026.05 | 59 | 69 | |
| AutoFactModel=Llama-3.2-1B2026.05 | 57 | 69 | |
| AutoFactModel=Mistral-7B-v0.32026.05 | 57 | 70 | |
| AutoFactModel=Qwen3-4B2026.05 | 57 | 70 | |
| AutoFactModel=Llama-3.1-8B2026.05 | 57 | 69 | |
| SEPModel=Llama-3.1-8B2026.05 | 46 | 67 |