# Token-level hallucination detection on RAGTruth Source: [Token-level hallucination detection on RAGTruth]() Collection index: [llms.txt]() Data updated: 2026-07-01T07:11:30.468Z Evaluation of model performance on detecting hallucinations specifically at the token level using the RAGTruth dataset, reporting AP and AUROC scores. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Token-level hallucination detection | | Dataset (including any stated split) | RAGTruth | | Metric | AP (Token-level) | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 18 of 18 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [AP (Token-level)](): Higher is better; unit: PERCENT. - [AUROC (Token-level)](): Higher is better; unit: PERCENT. ## Recorded results | Method | AP (Token-level) | Result date | Report paper | | --- | --- | --- | --- | | CORTEX | 59.4 | Not recorded | 6a449f02e0224bd4f4992fc1 | | CORTEX | 56.86 | Not recorded | 6a449f02e0224bd4f4992fc1 | | CORTEX | 54.73 | Not recorded | 6a449f02e0224bd4f4992fc1 | | RAGLens | 45.75 | Not recorded | 6a449f02e0224bd4f4992fc1 | | RAGLens | 42.71 | Not recorded | 6a449f02e0224bd4f4992fc1 | | RAGLens | 42.44 | Not recorded | 6a449f02e0224bd4f4992fc1 | | SAPLMA | 41.82 | Not recorded | 6a449f02e0224bd4f4992fc1 | | SAPLMA | 40.27 | Not recorded | 6a449f02e0224bd4f4992fc1 | | SAPLMA | 38.94 | Not recorded | 6a449f02e0224bd4f4992fc1 | | ICR Probe | 26.65 | Not recorded | 6a449f02e0224bd4f4992fc1 | | ICR Probe | 26.21 | Not recorded | 6a449f02e0224bd4f4992fc1 | | ICR Probe | 24.64 | Not recorded | 6a449f02e0224bd4f4992fc1 | | LLM-Check | 9.51 | Not recorded | 6a449f02e0224bd4f4992fc1 | | LLM-Check | 7.69 | Not recorded | 6a449f02e0224bd4f4992fc1 | | LLM-Check | 7.43 | Not recorded | 6a449f02e0224bd4f4992fc1 | | NLL | 4.95 | Not recorded | 6a449f02e0224bd4f4992fc1 | | NLL | 4.79 | Not recorded | 6a449f02e0224bd4f4992fc1 | | NLL | 4.76 | Not recorded | 6a449f02e0224bd4f4992fc1 | ### CORTEX AP (Token-level): 59.4. Result ID: 6a44bda173d097d845be1b71. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen"}. This record does not assert independent reproduction. ### CORTEX AP (Token-level): 56.86. Result ID: 6a44bda173d097d845be1b70. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama"}. This record does not assert independent reproduction. ### CORTEX AP (Token-level): 54.73. Result ID: 6a44bda173d097d845be1b72. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Mistral"}. This record does not assert independent reproduction. ### RAGLens AP (Token-level): 45.75. Result ID: 6a44bda173d097d845be1b6e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen"}. This record does not assert independent reproduction. ### RAGLens AP (Token-level): 42.71. Result ID: 6a44bda173d097d845be1b6d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama"}. This record does not assert independent reproduction. ### RAGLens AP (Token-level): 42.44. Result ID: 6a44bda173d097d845be1b6f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Mistral"}. This record does not assert independent reproduction. ### SAPLMA AP (Token-level): 41.82. Result ID: 6a44bda173d097d845be1b6b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen"}. This record does not assert independent reproduction. ### SAPLMA AP (Token-level): 40.27. Result ID: 6a44bda173d097d845be1b6c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Mistral"}. This record does not assert independent reproduction. ### SAPLMA AP (Token-level): 38.94. Result ID: 6a44bda173d097d845be1b6a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama"}. This record does not assert independent reproduction. ### ICR Probe AP (Token-level): 26.65. Result ID: 6a44bda173d097d845be1b67. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama"}. This record does not assert independent reproduction. ### ICR Probe AP (Token-level): 26.21. Result ID: 6a44bda173d097d845be1b69. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Mistral"}. This record does not assert independent reproduction. ### ICR Probe AP (Token-level): 24.64. Result ID: 6a44bda173d097d845be1b68. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen"}. This record does not assert independent reproduction. ### LLM-Check AP (Token-level): 9.51. Result ID: 6a44bda173d097d845be1b65. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen"}. This record does not assert independent reproduction. ### LLM-Check AP (Token-level): 7.69. Result ID: 6a44bda173d097d845be1b64. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Llama"}. This record does not assert independent reproduction. ### LLM-Check AP (Token-level): 7.43. Result ID: 6a44bda173d097d845be1b66. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Mistral"}. This record does not assert independent reproduction. ### NLL AP (Token-level): 4.95. Result ID: 6a44bda173d097d845be1b62. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen"}. This record does not assert independent reproduction. ### NLL AP (Token-level): 4.79. Result ID: 6a44bda173d097d845be1b63. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Mistral"}. This record does not assert independent reproduction. ### NLL AP (Token-level): 4.76. Result ID: 6a44bda173d097d845be1b61. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.