ResearchBenchmarksHallucination Detection on RAGTruth Span-level conflict vs cleanFollow70.4AUCGASP68.00868.62969.2569.871Jul 5, 2026Evaluation ResultsMethodMethodLinksAUCGASPScorer=Qwen2.5-1.5B, P...Scorer=Qwen2.5-1.5B, Protocol=leakage-clean2026.0770.4GASPScorer=Qwen2.5-0.5B, P...Scorer=Qwen2.5-0.5B, Protocol=leakage-clean2026.0768.5GASPScorer=SmolLM2-1.7B, P...Scorer=SmolLM2-1.7B, Protocol=leakage-clean2026.0768.1