Hallucination benchmarks are all over the place in 2026. Depending on which...

https://foxtrot-wiki.win/index.php/How_to_Detect_Hallucinations_When_the_Answer_Sounds_Polished_and_Final

Hallucination benchmarks are all over the place in 2026. Depending on which test you run, your model's accuracy shifts wildly. For instance, the HalluHard benchmark shows a 30.2% error rate even with web search enabled