ResearchBenchmarksGeneral AI Assistant Tasks on GAIA Level 3 (val)Follow44.2AccuracyGPT-55.09615.24825.435.552Dec 7, 2025Evaluation ResultsMethodMethodLinksAccuracy95% Confidence IntervalBetween-Subject VarianceIntraclass Correlation Coefficient (ICC)GPT-5Model=GPT-5, Web searc...Model=GPT-5, Web search=true, Trials per question=642025.1244.2—0.160.629GPT-4oModel=GPT-4o, Web sear...Model=GPT-4o, Web search=true, Trials per question=642025.126.6—0.0190.304