Data as of Jul 25, 2026 · Based on 2,718,867 AI responses across 9,511 prompts · See how Parse measures this
RAGTruthgithub.com · Low evidenceRAGTruth is a word-level hallucination corpus designed for training and evaluating large language models within retrieval-augmented generation (RAG) settings. It contains nearly 18,000 manually annotated responses from diverse LLMs, with detailed labels for hallucination spans and intensity.
Parse Score