Are benchmarks finally getting honest about AI hallucinations? By 2026, rates...

https://garrettwigp625.tearosediner.net/why-did-vectara-hallucination-rates-jump-on-the-7-700-document-dataset

Are benchmarks finally getting honest about AI hallucinations? By 2026, rates vary wildly depending on the test used. HalluHard now shows a 30.2% failure rate even with web search enabled