Hallucination rates depend on the yardstick. Whether you measure via Vectara...

https://spark-wiki.win/index.php/Which_Benchmark_Should_You_Cite_for_Multi-Turn_Chat_Apps_with_Citations%3F

Hallucination rates depend on the yardstick. Whether you measure via Vectara HHEM or HalluHard—where models hit 30.2% failure—results vary wildly. In 2026, don’t ask if your AI is accurate; ask which benchmark fits your use case