Accuracy benchmarks in 2026 are inconsistent. Hallucination rates swing wildly...
https://station-wiki.win/index.php/GPT-5.2-thinking_with_Web_Search_Still_Hit_38.2%25_Hallucination:_Why_So_High%3F
Accuracy benchmarks in 2026 are inconsistent. Hallucination rates swing wildly between tests, like the 30.2% error rate on HalluHard with web search