Hallucination benchmarks are all over the map. Rates vary wildly by test, and...

https://dibz.me/blog/gemini-2-0-flash-001-at-0-7-hallucination-rate-why-your-production-pipeline-needs-a-reality-check-1160

Hallucination benchmarks are all over the map. Rates vary wildly by test, and HalluHard shows a 30.2% error rate even with web search enabled. Relying on a single metric is a bad bet for your production stack