No need benchmarks. We already know they can BS better than anyone for 3 hours, make statistical method errors, and hallucinate studies.