ResearchBenchmarksTool-use Question Answering on HotpotQA 500 samples (evaluation)Follow71.6AccuracyBaseline70.97671.13871.371.462May 13, 2026Evaluation ResultsMethodMethodLinksAccuracyLatencySpeedupBaselineModel=openai-realtime-1.5Model=openai-realtime-1.52026.0571.64.5—AsyncIOModel=openai-realtime-1.5Model=openai-realtime-1.52026.05713.61.3