List your company

Hallucination Detection on HaluBench (AUROC)

97AUROCF: Ans. Expect.
-3.3995222.6657448.73174.79626May 16, 2026

Evaluation results

MethodLinks
F: Ans. Expect.Backbone=Llama-3.3-70B2026.0597
Ensemble A+C+E+FBackbone=Llama-3.3-70B2026.0596
E: DRIFTBackbone=Llama-3.3-70B2026.0591
DRIFT-concat (ablation of E)Backbone=Llama-3.3-70B2026.0591
SAPLMABackbone=Llama-3.3-70B2026.0591
HaMIBackbone=Llama-3.3-70B2026.0587
D: Act. VarianceBackbone=Llama-3.3-70B2026.0582
PRISMBackbone=Llama-3.3-70B2026.0580
TXTEMB (ctrl)Backbone=Llama-3.3-70B2026.0577
DRIFT-logpBackbone=Llama-3.3-70B2026.0576
A: PerturbationBackbone=Llama-3.3-70B2026.0568
IRIS⋆ (self-verify)Backbone=Llama-3.3-70B2026.0567
MINDBackbone=Llama-3.3-70B2026.0567
HaloScopeBackbone=Llama-3.3-70B2026.0567
SEPsBackbone=Llama-3.3-70B2026.0567
C: Sem. EntropyBackbone=Llama-3.3-70B2026.0562
DoLaBackbone=Llama-3.3-70B2026.0562
ACT (detect)Backbone=Llama-3.3-70B2026.0562
P(True)Backbone=Llama-3.3-70B2026.0560
Log-ProbBackbone=Llama-3.3-70B2026.0556
SelfCheckGPT-NLIBackbone=Llama-3.3-70B2026.0554
INSIDEBackbone=Llama-3.3-70B2026.0550
HalluShiftBackbone=Llama-3.3-70B2026.0531
SAPLMAModel=Gemma-3-12b2026.050.935
DRIFT-concatModel=Gemma-3-12b2026.050.928
DRIFTModel=Llama-3.3-70B2026.050.915
DRIFTModel=Gemma-3-12b2026.050.913
SAPLMAModel=Llama-3.3-70B2026.050.911
DRIFT-concatModel=Llama-3.3-70B2026.050.911
DRIFT-concatModel=Mistral-24B2026.050.91
DRIFT-concatModel=Gemma-3-4b2026.050.904
DRIFTModel=Qwen2.5-72B2026.050.9
DRIFTModel=Mistral-24B2026.050.9
SAPLMAModel=Mistral-24B2026.050.9
DRIFTModel=Phi-4-mini2026.050.9
DRIFT-concatModel=Phi-4-mini2026.050.9
DRIFT-concatModel=Qwen2.5-72B2026.050.898
DRIFT-concatModel=Mixtral-8x7B2026.050.898
DRIFTModel=Gemma-3-4b2026.050.896
SAPLMAModel=Gemma-3-4b2026.050.895
SAPLMAModel=Qwen3-8B2026.050.894
SAPLMAModel=Mixtral-8x7B2026.050.893
SAPLMAModel=Phi-4-mini2026.050.891
DRIFTModel=Mixtral-8x7B2026.050.889
SAPLMAModel=Qwen2.5-72B2026.050.885
DRIFT-concatModel=Qwen3-32B2026.050.883
DRIFT-concatModel=Mean2026.050.883
DRIFTModel=Qwen3-32B2026.050.88
DRIFT-concatModel=Qwen3-8B2026.050.879
DRIFTModel=Mean2026.050.879
SAPLMAModel=Mean2026.050.879
DRIFT-concatModel=DS-R1-Llama-70B2026.050.873
DRIFTModel=DS-R1-Llama-70B2026.050.871
DRIFTModel=Qwen3-8B2026.050.868
SAPLMAModel=DS-R1-Llama-70B2026.050.865
SAPLMAModel=Qwen3-32B2026.050.855
DRIFTModel=Llama-3.1-8B2026.050.839
SAPLMAModel=Llama-3.1-8B2026.050.824
DRIFT-concatModel=Llama-3.1-8B2026.050.823
SAPLMAModel=DS-R1-Qwen-7B2026.050.792
DRIFT-concatModel=DS-R1-Qwen-7B2026.050.788
DRIFTModel=DS-R1-Qwen-7B2026.050.774
SelfCheckGPT-NLIModel=Qwen3-32B, Passes=102026.050.679
SelfCheckGPT-NLIModel=Qwen3-8B, Passes=102026.050.678
SelfCheckGPT-NLIModel=Gemma-3-4b, Passes=102026.050.637
SelfCheckGPT-NLIModel=DS-R1-Llama-70B, Passes=102026.050.635
SelfCheckGPT-NLIModel=Qwen2.5-72B, Passes=102026.050.613
SelfCheckGPT-NLIModel=Mixtral-8x7B, Passes=102026.050.607
SelfCheckGPT-NLIModel=Llama-3.1-8B, Passes=102026.050.579
SelfCheckGPT-NLIModel=Mistral-24B, Passes=102026.050.564
SelfCheckGPT-NLIModel=Mean, Passes=102026.050.557
SelfCheckGPT-NLIModel=Llama-3.3-70B, Passes=102026.050.535
SelfCheckGPT-NLIModel=DS-R1-Qwen-7B, Passes=102026.050.524
SelfCheckGPT-NLIModel=Phi-4-mini, Passes=102026.050.48
SelfCheckGPT-NLIModel=Gemma-3-12b, Passes=102026.050.462