Hallucination Detection on RAGTruth (test)
0.9096AUROCTPA
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| TPAModel=Mistral-7B2025.12 | 0.9096 | 87.02 | — | 92 | — | — | |
| TPABase Model=Mistral-7B2025.12 | 0.9096 | 87.02 | — | 92 | — | — | |
| TPAModel=LLaMA2-13B2025.12 | 0.8681 | 79.75 | — | 79.13 | — | — | |
| SpikeScoreLLM Backbone=Qwen3-14B, Method Category=Cross-domain specialized methods2026.01 | 0.8535 | — | — | — | — | — | |
| NovoModel=LLaMA2-13B2025.12 | 0.8506 | 77.33 | — | 78.26 | — | — | |
| NovoModel=Mistral-7B2025.12 | 0.8419 | 80 | — | 87.65 | — | — | |
| NovoBase Model=Mistral-7B2025.12 | 0.8419 | 80 | — | 87.65 | — | — | |
| RagtStack+calAnalyzer=Gemma-2-9B2026.05 | 0.836 | 71.3 | — | — | 74.7 | 0.103 | |
| RagtStack+calAnalyzer=Qwen2.5-7B2026.05 | 0.834 | 71.7 | — | — | 75.3 | 0.101 | |
| SpikeScoreLLM Backbone=Qwen3-8B, Method Category=Cross-domain specialized methods2026.01 | 0.8291 | — | — | — | — | — | |
| LettuceAccess level=Black Box2026.02 | 0.8264 | 0.745 | 80 | — | — | — | |
| NovoModel=LLaMA3-8B2025.12 | 0.8258 | 78.01 | — | 77.37 | — | — | |
| RagtStack rawAnalyzer=Qwen2.5-0.5B2026.05 | 0.825 | 70 | — | — | 74.7 | 0.093 | |
| ReDeEPModel=LLaMA2-13B2025.12 | 0.8244 | 75.87 | — | 71.98 | — | — | |
| RagtStack+calAnalyzer=LLaMA-3-3B2026.05 | 0.822 | 69.8 | — | — | 74.3 | 0.09 | |
| TPAModel=LLaMA3-8B2025.12 | 0.8211 | 78.43 | — | 78.6 | — | — | |
| RagtStack rawAnalyzer=LLaMA-3-8B2026.05 | 0.819 | 68.7 | — | — | 73.5 | 0.086 | |
| RagtStack+calAnalyzer=Pythia-1.4B2026.05 | 0.818 | 69.2 | — | — | 73.8 | 0.086 | |
| Stacking+calAnalyzer=Gemma-2-9B2026.05 | 0.816 | 69.3 | — | — | 72.8 | 0.084 | |
| SpikeScoreLLM Backbone=Llama3.1-8B, Method Category=Cross-domain specialized methods2026.01 | 0.8154 | — | — | — | — | — | |
| RagtStack+calAnalyzer=Gemma-2-2B2026.05 | 0.814 | 67.2 | — | — | 72.4 | 0.081 | |
| TSVModel=LLaMA2-13B2025.12 | 0.8123 | 69.87 | — | 80.68 | — | — | |
| Stacking+calAnalyzer=Qwen2.5-7B2026.05 | 0.811 | 68.4 | — | — | 73.4 | 0.079 | |
| Stacking+calAnalyzer=LLaMA-3-3B2026.05 | 0.809 | 68.1 | — | — | 73.6 | 0.077 | |
| SEPModel=LLaMA2-13B2025.12 | 0.8089 | 71.59 | — | 65.8 | — | — | |
| Stacking+calAnalyzer=LLaMA-3-8B2026.05 | 0.806 | 66.9 | — | — | 72.3 | 0.074 | |
| ITIModel=LLaMA2-13B2025.12 | 0.8051 | 68.38 | — | 55.19 | — | — | |
| Stacking+calAnalyzer=Qwen2.5-0.5B2026.05 | 0.803 | 68.2 | — | — | 72.1 | 0.07 | |
| Stacking+calAnalyzer=Pythia-1.4B2026.05 | 0.802 | 64.5 | — | — | 71.4 | 0.07 | |
| Stacking+calAnalyzer=Gemma-2-2B2026.05 | 0.798 | 67.9 | — | — | 71.6 | 0.065 | |
| TSVModel=Mistral-7B2025.12 | 0.7972 | 67.64 | — | 55.38 | — | — | |
| TSVBase Model=Mistral-7B2025.12 | 0.7972 | 67.64 | — | 55.38 | — | — | |
| TPAModel=LLaMA2-7B2025.12 | 0.7873 | 72.38 | — | 83.28 | — | — | |
| RefCheckModel=LLaMA2-13B2025.12 | 0.7857 | 70.23 | — | 68 | — | — | |
| TSVModel=LLaMA3-8B2025.12 | 0.7769 | 64.42 | — | 55.46 | — | — | |
| ICR ProbeLLM Backbone=Qwen3-14B, Method Category=Cross-domain specialized methods2026.01 | 0.7653 | — | — | — | — | — | |
| SpikeScoreLLM Backbone=Llama3.2-3B, Method Category=Cross-domain specialized methods2026.01 | 0.7631 | — | — | — | — | — | |
| NovoModel=LLaMA2-7B2025.12 | 0.7608 | 70.57 | — | 82.74 | — | — | |
| RAGASModel=LLaMA2-13B2025.12 | 0.7541 | 67.47 | — | 67.63 | — | — | |
| ReDeEP, chunk level2026.05 | 0.746 | 69.5 | — | — | — | 0.013 | |
| ReDeEPModel=LLaMA2-7B2025.12 | 0.7458 | 71.9 | — | 80.97 | — | — | |
| ICR ProbeLLM Backbone=Qwen3-8B, Method Category=Cross-domain specialized methods2026.01 | 0.7426 | — | — | — | — | — | |
| ICR ProbeLLM Backbone=Llama3.2-3B, Method Category=Cross-domain specialized methods2026.01 | 0.7416 | — | — | — | — | — | |
| ChainPollModel=LLaMA2-13B2025.12 | 0.7414 | 73.42 | — | 78.74 | — | — | |
| ICR ProbeLLM Backbone=Llama3.1-8B, Method Category=Cross-domain specialized methods2026.01 | 0.7405 | — | — | — | — | — | |
| ReDeEP, token level2026.05 | 0.733 | 71.3 | — | — | — | — | |
| RAGASModel=LLaMA2-7B2025.12 | 0.729 | 66.67 | — | 63.27 | — | — | |
| ReDeEPModel=LLaMA3-8B2025.12 | 0.7285 | 69.47 | — | 78.19 | — | — | |
| ITIModel=LLaMA2-7B2025.12 | 0.7161 | 67.45 | — | 54.16 | — | — | |
| SEPModel=LLaMA2-7B2025.12 | 0.7143 | 66.27 | — | 74.77 | — | — | |
| TrulensModel=LLaMA2-13B2025.12 | 0.7073 | 68.67 | — | 77.29 | — | — | |
| HALT-LAccess level=Black Box2026.02 | 0.7065 | 0.59 | 65.7 | — | — | — | |
| SEPModel=LLaMA3-8B2025.12 | 0.7004 | 69.15 | — | 73.33 | — | — | |
| EigenScoreLLM Backbone=Llama3.1-8B, Method Category=Training-free methods2026.01 | 0.6974 | — | — | — | — | — | |
| Lexical SimilarityLLM Backbone=Llama3.1-8B, Method Category=Training-free methods2026.01 | 0.6958 | — | — | — | — | — | |
| RefCheckModel=LLaMA2-7B2025.12 | 0.6912 | 67.36 | — | 62.8 | — | — | |
| EigenScoreLLM Backbone=Qwen3-14B, Method Category=Training-free methods2026.01 | 0.6904 | — | — | — | — | — | |
| Semantic EntropyLLM Backbone=Qwen3-14B, Method Category=Training-free methods2026.01 | 0.6796 | — | — | — | — | — | |
| RAGASModel=LLaMA3-8B2025.12 | 0.6776 | 50.94 | — | 39.09 | — | — | |
| ChainPollModel=LLaMA2-7B2025.12 | 0.6738 | 70.66 | — | 78.32 | — | — | |
| ChainPollModel=LLaMA3-8B2025.12 | 0.6687 | 58.13 | — | 44.86 | — | — | |
| EigenScoreModel=LLaMA2-13B2025.12 | 0.664 | 66.37 | — | 67.15 | — | — | |
| H_altsAccess level=Aggregate Statistics2026.02 | 0.663 | 0.5398 | 37.24 | — | — | — | |
| TSVModel=LLaMA2-7B2025.12 | 0.6609 | 66.32 | — | 55.26 | — | — | |
| Delta H_decAccess level=Aggregate Statistics2026.02 | 0.6565 | 0.5624 | 47.3 | — | — | — | |
| ITIModel=LLaMA3-8B2025.12 | 0.6534 | 69.33 | — | 68.5 | — | — | |
| TrulensModel=LLaMA2-7B2025.12 | 0.651 | 65.67 | — | 68.14 | — | — | |
| EigenScoreModel=LLaMA3-8B2025.12 | 0.6497 | 67.45 | — | 70.78 | — | — | |
| Lexical SimilarityLLM Backbone=Qwen3-14B, Method Category=Training-free methods2026.01 | 0.6479 | — | — | — | — | — | |
| TrulensModel=LLaMA3-8B2025.12 | 0.6464 | 50.53 | — | 39.09 | — | — | |
| Lexical SimilarityLLM Backbone=Qwen3-8B, Method Category=Training-free methods2026.01 | 0.644 | — | — | — | — | — | |
| PerplexityLLM Backbone=Qwen3-14B, Method Category=Training-free methods2026.01 | 0.642 | — | — | — | — | — | |
| Semantic EntropyLLM Backbone=Qwen3-8B, Method Category=Training-free methods2026.01 | 0.6388 | — | — | — | — | — | |
| SAPLMALLM Backbone=Qwen3-14B, Method Category=Training-based methods2026.01 | 0.6338 | — | — | — | — | — | |
| EigenScoreLLM Backbone=Llama3.2-3B, Method Category=Training-free methods2026.01 | 0.6329 | — | — | — | — | — | |
| PPLAccess level=Aggregate Statistics2026.02 | 0.6302 | 0.518 | 26.01 | — | — | — | |
| EigenScoreLLM Backbone=Qwen3-8B, Method Category=Training-free methods2026.01 | 0.6292 | — | — | — | — | — | |
| PerplexityLLM Backbone=Qwen3-8B, Method Category=Training-free methods2026.01 | 0.6284 | — | — | — | — | — | |
| Semantic EntropyLLM Backbone=Llama3.2-3B, Method Category=Training-free methods2026.01 | 0.621 | — | — | — | — | — | |
| MMLLM Backbone=Qwen3-14B, Method Category=Training-based methods2026.01 | 0.6191 | — | — | — | — | — | |
| Lexical SimilarityLLM Backbone=Llama3.2-3B, Method Category=Training-free methods2026.01 | 0.6179 | — | — | — | — | — | |
| PerplexityLLM Backbone=Llama3.1-8B, Method Category=Training-free methods2026.01 | 0.6151 | — | — | — | — | — | |
| EigenScoreModel=LLaMA2-7B2025.12 | 0.6045 | 66.82 | — | 74.69 | — | — | |
| MMLLM Backbone=Qwen3-8B, Method Category=Training-based methods2026.01 | 0.6044 | — | — | — | — | — | |
| Semantic EntropyLLM Backbone=Llama3.1-8B, Method Category=Training-free methods2026.01 | 0.6037 | — | — | — | — | — | |
| RefCheckModel=LLaMA3-8B2025.12 | 0.6014 | 46.28 | — | 35.8 | — | — | |
| SAPLMALLM Backbone=Llama3.1-8B, Method Category=Training-based methods2026.01 | 0.6011 | — | — | — | — | — | |
| PerplexityLLM Backbone=Llama3.2-3B, Method Category=Training-free methods2026.01 | 0.5936 | — | — | — | — | — | |
| SAPLMALLM Backbone=Qwen3-8B, Method Category=Training-based methods2026.01 | 0.5904 | — | — | — | — | — | |
| SAPLMALLM Backbone=Llama3.2-3B, Method Category=Training-based methods2026.01 | 0.5874 | — | — | — | — | — | |
| MMLLM Backbone=Llama3.1-8B, Method Category=Training-based methods2026.01 | 0.5839 | — | — | — | — | — | |
| LLM-Check Attn ScoreAccess level=White Box2026.02 | 0.583 | 0.5718 | — | — | — | — | |
| LLM-Check Hidden ScoreAccess level=White Box2026.02 | 0.5724 | 0.4745 | — | — | — | — | |
| MMLLM Backbone=Llama3.2-3B, Method Category=Training-based methods2026.01 | 0.5587 | — | — | — | — | — | |
| SEPLLM Backbone=Llama3.1-8B, Method Category=Training-based methods2026.01 | 0.5547 | — | — | — | — | — | |
| SEPLLM Backbone=Qwen3-8B, Method Category=Training-based methods2026.01 | 0.5253 | — | — | — | — | — | |
| SEPLLM Backbone=Qwen3-14B, Method Category=Training-based methods2026.01 | 0.5239 | — | — | — | — | — | |
| SEPLLM Backbone=Llama3.2-3B, Method Category=Training-based methods2026.01 | 0.4988 | — | — | — | — | — | |
| H_overallAccess level=Aggregate Statistics2026.02 | 0.3698 | 0.5177 | 25.89 | — | — | — |