Hallucination rates fluctuate wildly between benchmarks, so don't trust the...

https://emilioezho292.theglensecret.com/gpt-5-2-thinking-with-web-search-still-hit-38-2-hallucination-why-so-high

Hallucination rates fluctuate wildly between benchmarks, so don't trust the marketing. With HalluHard showing a 30.2% error rate despite web search, you need better tests. Focus on metrics that mirror your production environment